Book of peer-reviewed papers 10th Congress of the Alps Adria Acoustics Association International Scientific Congress 20. – 21. September 2023, Izola, Slovenia 2 Book of peer-reviewed papers. 10th Congress of the Alps Adria Acoustics Association, 20. – 21. September 2023, Izola, Slovenia. International Scientific Congress Editor-in-chief Assoc. Prof. Dr. Mateja Dovjak, Congress Chair, President of the Slovenian Acoustical Society (SDA), Faculty of Civil and Geodetic Engineering of the University of Ljubljana, Slovenia Editorial board Assoc. Prof. Dr. Mateja Dovjak, Congress Chair, President of the Slovenian Acoustical Society (SDA), Faculty of Civil and Geodetic Engineering of the University of Ljubljana, Slovenia Assist. Prof. Dr. Rok Prislan, Secretary of the SDA, InnoRenew CoE, Slovenia Andrej Biček, M.Phil., Treasurer of the SDA, Nela d.o.o, Slovenia Assist. Prof. Dr. Samo Beguš, University of Ljubljana, Faculty of Electrical Engineering, Slovenia Nika Šubic, M.Sc., MK3 d.o.o., Slovenia Luka Čurović, M.Sc., University of Ljubljana, Faculty of Mechanical Engineering, Slovenia Assist. Prof. Dr. Teja Povh, Faculty of Civil and Geodetic Engineering of the University of Ljubljana, Slovenia Organisational board Assoc. Prof. Dr. Mateja Dovjak, Congress Chair, President of the Slovenian Acoustical Society (SDA), Faculty of Civil and Geodetic Engineering of the University of Ljubljana, Slovenia Assist. Prof. Dr. Rok Prislan, Secretary of the SDA, InnoRenew CoE, Slovenia Andrej Biček, M.Phil., Treasurer of the SDA, Nela d.o.o, Slovenia Assist. Prof. Dr. Samo Beguš, University of Ljubljana, Faculty of Electrical Engineering, Slovenia Dr. Egon Susič, Danfoss Trata d.o.o., Slovenia Nika Šubic, M.Sc., MK3 d.o.o., Slovenia Luka Čurović, M.Sc., University of Ljubljana, Faculty of Mechanical Engineering, Slovenia Klara Rupnik, M.Sc., IMS Merilni sistemi d.o.o., Slovenia Prof. Dr. Mirko Čudina, founder of SDA and one of the founders of the AAAA and EUROREGIO, University of Ljubljana, Faculty of Mechanical Engineering, Slovenia Špela Cenček, M.Phil., Marbo Okolje d.o.o., Slovenia Prof. Dr. Tino Bucak, HAD, University of Zagreb, Faculty of Transport and Traffic Sciences, Croatia Prof. Dr. Kristian Jambrošić, University of Zagreb Faculty of Electrical Engineering and Computing, Croatia Beáta Mesterházy, M.Sc., President of the Department of Acoustics of OPAKFI, BME (Budapest University of Tecnology and Economics), Department of Building Constructions, Laboratory of Building Acoustics András Muntag, M.Sc., President of the Department of Noise and vibration control of OPAKFI, Enviroplus Kft. (Ltd), Hungary Assist. Prof. Dr. Jurica Ivošević, University of Zagreb, Department of Aeronautics, Croatia Attila Balázs Nagy, M.Sc., BME (Budapest University of Technology and Economics), Laboratory of Building Acoustics, Hungary Prof. Dr. Antonio Petošić, University of Zagreb, Faculty of Electrical Engineering and Computing, Croatia Scientific board, reviewers Assoc. Prof. Dr. Mateja Dovjak, Congress Chair, President of the Slovenian Acoustical Society (SDA), Faculty of Civil and Geodetic Engineering of the University of Ljubljana, Slovenia Assist. Prof. Dr. Rok Prislan, Secretary of the SDA, InnoRenew CoE, Slovenia Andrej Biček, M.Phil., Treasurer of the SDA, Nela d.o.o, Slovenia Assist. Prof. Dr. Samo Beguš, University of Ljubljana, Faculty of Electrical Engineering, Slovenia Dr. Egon Susič, Danfoss Trata d.o.o., Slovenia 3 Prof. Dr. Mirko Čudina, founder of SDA and one of the founders of the AAAA and EUROREGIO, University of Ljubljana, Faculty of Mechanical Engineering, Slovenia Dr. Ferdinand Deželak, a retired researcher Prof. Dr. Daniel Svenšek, University of Ljubljana, Faculty of Mathematics and Physics, Slovenia Dr. Jernej Polajnar, National Institute of Biology, Slovenia Assoc. Prof. Dr. Jurij Prezelj, University of Ljubljana, Faculty of Mechanical Engineering, Slovenia Assoc. Prof. Dr. Nikola Holeček, University of Ljubljana, Faculty of Chemistry and Chemical Technology, Faculty of Environmental Protection, Slovenia Prof. Dr. Tino Bucak, HAD, University of Zagreb, Faculty of Transport and Traffic Sciences, Croatia Prof. Dr. Kristian Jambrošić, University of Zagreb Faculty of Electrical Engineering and Computing, Croatia Beáta Mesterházy, M.Sc., President of the Department of Acoustics of OPAKFI, BME (Budapest University of Tecnology and Economics), Department of Building Constructions, Laboratory of Building Acoustics András Muntag, M.Sc., President of the Department of Noise and vibration control of OPAKFI, Enviroplus Kft. (Ltd), Hungary Prof. Dr. Marko Horvat, President of the HAD, Faculty of Electrical Engineering and Computing, University of Zagreb Prof. Dr. Antonio Petošić, University of Zagreb, Faculty of Electrical Engineering and Computing, University of Zagreb Assist. Prof. Dr. Jurica Ivošević, University of Zagreb, Faculty of Transport and Traffic Sciences, Croatia Attila Balázs Nagy, M.Sc., BME (Budapest University of Technology and Economics), Laboratory of Building Acoustics, Hungary Viktor Jemec, M.Phil., Društvo vzdrževalcev Slovenije (DVS), SDA Bibliographic information support: Assist. Prof. Dr. Teja Povh, Faculty of Civil and Geodetic Engineering of the University of Ljubljana, Slovenia Page layout cover page: Gertrud Fábián, Information Technology Support, InnoRenew CoE Publisher: Slovensko društvo za akustiko, Slovenian Acoustical Society (SDA), Jamova cesta 2, 1000 Ljubljana, Slovenia Place and year: Ljubljana, 223 E-format available at URL: https://www.alpsadriaacoustics.eu/ The Book of peer-reviewed abstracts was made as part of the 1th Congress of the Alps Adria Acoustics Association, 2. – 21. September 223, Izola, Slovenia. Papers are peer-reviewed. Notice The publisher assumes no responsibility for any injury and/or damage to persons or property due to product liability, negligence or otherwise, or due to the use or operation of any methods, products, instructions or ideas contained in the material herein. The final layout and the content of the manuscript is the sole responsibility of the authors. 4 Copyright © by Slovensko društvo za akustiko, Slovenian Acoustical Society (SDA), Jamova cesta 2, 1000 Ljubljana, Slovenia. All rights reserved. Reproduction and distribution under the copyright law are not allowed! Kataložni zapis o publikaciji (CIP) pripravili v Narodni in univerzitetni knjižnici v Ljubljani COBISS.SI-ID 173997315 ISBN 978-961-94085-2-0 (PDF) URL: https://www.alpsadriaacoustics.eu/ 5 Organizing Institution Slovensko društvo za akustiko, Slovenian Acoustical Society (SDA) Jamova cesta 2, 1000 Ljubljana, Slovenia Contact Congress Secretariat E-mail: info@alpsadriaacoustics.eu Postal address: Slovensko društvo za akustiko, Slovenian Acoustical Society (SDA) Jamova cesta 2, 1000 Ljubljana, Slovenia Organizers Slovensko društvo za akustiko, Slovenian Acoustical Society (SDA) InnoRenew CoE University of Ljubljana, Faculty of Civil and Geodetic Engineering (UL FGG) Co-organisers Alps Adria Acoustics Association (AAAA) Hrvatsko akutsičko društvo, Acoustical Society of Croatia (HAD) Hungarian Scientific Society for Optics, Acoustics, Motion Picture and Theatre Technology (OPAKFI) European Acoustics Association (EAA) Gold sponsors IMS Merilni sistemi d.o.o. Rothoblaas DEWESoft Knaufinsulation Silver sponsors HEAD acoustics Getzner Bronze sponsors Fragmat Domel Ursa 6 Contents Preface ------------------------------------------------------------------------------------------------------------------------------------- 7 Program overview ---------------------------------------------------------------------------------------------------------------------- 8 Program --------------------------------------------------------------------------------------------------------------------------------- 10 Keynote ---------------------------------------------------------------------------------------------------------------------------------- 16 Prof. Dr. Janko Slavič: High-speed camera based identification of sound and vibrations Contributed papers: Room acoustics ---------------------------------------------------------------------------------------------- 17 Contributed papers: Noise and vibrations --------------------------------------------------------------------------------------- 68 Keynote -------------------------------------------------------------------------------------------------------------------------------- 115 Prof. Dr.-Ing. Roland Sottek: Development of sound quality metrics using models based on human perception and their applications Contributed papers: Advanced measurement techniques in acoustics --------------------------------------------------- 117 Contributed papers: Acoustic software and training ------------------------------------------------------------------------- 153 Keynote -------------------------------------------------------------------------------------------------------------------------------- 174 Prof. Dr. Jonas Brunskog: Challenges in sound insulation of wooden buildings Technical keynote ------------------------------------------------------------------------------------------------------------------- 175 Assist Prof. Dr. Rok Prislan: Design and construction of the InnoRenew CoE Acoustics Laboratory Contributed papers: Building acoustics ------------------------------------------------------------------------------------------ 176 Contributed papers: Soundscape and sound reproduction techniques -------------------------------------------------- 196 Contributed papers: Noise and vibrations -------------------------------------------------------------------------------------- 208 Keynote -------------------------------------------------------------------------------------------------------------------------------- 237 Prof. Dr. Goran Pavić: Modelling of sound and vibration using a virtual-source approach Contributed papers: Soundscape and sound reproduction techniques -------------------------------------------------- 238 Contributed papers: Noise and vibrations -------------------------------------------------------------------------------------- 249 Contributed papers: Advanced meausrement techniques in acoustics --------------------------------------------------- 274 Index--------------------------------------------------------------------------------------------------------------------------------------- 298 7 Preface Welcome to the 10th Congress of the Alps Adria Acoustics Association! The 10th Congress of the Alps Adria Acoustics Association will be held in Slovenia, at InnoRenew CoE in Izola, from 20th to 21st September 2023. The Alps Adria Acoustics Association (AAAA) was founded by the acoustics societies of Slovenia, Croatia and Austria in 2002 as a new regional association. In 2019 the Hungarian Scientific Society for Optics, Acoustics, Motion Picture and Theatre Technology (OPAKFI) joined instead of Austria. The original goal of AAAA was to promote all aspects of research in the field of acoustics in the region. In addition, the Association’s aim was to improve the overall cooperation among the countries and their respective national societies. Every two years, one of the three member societies of the AAAA organizes a scientific congress on acoustics. The main goal of these congresses is to bring together acousticians from Croatia, Hungary and Slovenia, as well as from the other European countries, in order to exchange knowledge, share research outcomes and strengthen mutual cooperation among these societies for the benefit of the whole region. The last event took place in Budapest in 2021. During the congress days, national and international experts present a number of scientific and applied papers on their research and professional activities in all fields of acoustics. In particular, the congress topics cover architectural and building acoustics, auditory and speech acoustic, environmental and transportation noise, machinery noise and vibration control, computational acoustics, electroacoustics, legislation in acoustics, musical acoustics, measurement techniques, non-linear acoustics, psychoacoustics and perception of sound, signal processing, sound generation and radiation, ultrasonics, hydroacoustics, etc. The scientific programme includes keynote lectures given by eminent international experts. The AAAA 2023 Congress deals with topics that are the focus of interest in the scientific community and among researchers working in the industry. The 2023 Congress is organized by the Slovenian Acoustical Society (SDA) acting on behalf of the Alps Adria Acoustics Association. The conference is planned to be a live conference. Looking forward to meeting you in Izola in September 2023! Assoc. Prof. Dr. Mateja Dovjak, Congress Chair, President of the Slovenian Acoustical Society (SDA), Faculty of Civil and Geodetic Engineering of the University of Ljubljana, Slovenia 8 Program overview WEDNESDAY, SEPTEMBER 20 TIME LOCATION REGISTRATION 8:00 - 9:00 SEQUOIA LECTURE ROOM OPENING CEREMONY 9:00 - 9:30 SEQUOIA LECTURE ROOM KEYNOTE Invited speech: 9:30 - 10:30 SEQUOIA LECTURE ROOM Prof. Dr. Janko Slavič COFFEE BREAK 10:30 - 10:50 MAIN HALL MORNING SESSIONS 10:50 - 12:50 SEQUOIA AND QUERCUS LECTURE ROOM LUNCH BUFFET 12:50 - 13:50 MAIN HALL ACOUSTIC LABORATORY VISIT - GROUP 1 13:50 - 14:10 ACOUSTIC LABORATORY INNORENEW COE GOLDEN SPONSOR PRESENTATION - 14:10 - 14:20 SEQUOIA LECTURE ROOM SPONSOR ROTHOBLAAS GOLDEN SPONSOR PRESENTATION - 14:20 - 14:30 SEQUOIA LECTURE ROOM SPONSOR KNAUF INSULATION KEYNOTE Invited speech: 14:30 - 15:30 SEQUOIA LECTURE ROOM Prof. Dr.-Ing. Roland Sottek COFFEE BREAK 15:30 - 15:50 MAIN HALL AFTERNOON SESSIONS 15:50 - 17:30 SEQUOIA AND QUERCUS LECTURE ROOM FREE TIME TRANSPORTATION TO DINNER 18:45 IN FRONT OF INNORENEW COE BUILDING GALA DINNER 19:00-22:00 RESTAURANT KAMIN 9 THURSDAY, SEPTEMBER 21 TIME LOCATION KEYNOTE Invited speech: 8:30 - 9:30 SEQUOIA LECTURE ROOM Prof. Dr. Jonas Brunskog TECHNICAL KEYNOTE Invited speech: 9:30 - 10:30 SEQUOIA LECTURE ROOM Assist Prof. Dr. Rok Prislan COFFEE BREAK 10:30 - 10:50 MAIN HALL MORNING SESSIONS 10:50 - 12:50 SEQUOIA AND QUERCUS LECTURE ROOM LUNCH BUFFET 12:50 - 13:50 MAIN HALL ACOUSTIC LABORATORY VISIT - GROUP 2 13:50 - 14:10 ACOUSTIC LABORATORY INNORENEW COE GOLDEN SPONSOR PRESENTATION - SPONSOR 14:10 - 14:20 SEQUOIA LECTURE ROOM DEWESOFT GOLDEN SPONSOR PRESENTATION - SPONSOR 14:20 - 14:30 SEQUOIA LECTURE ROOM IMS MERILNI SISTEMI D.O.O. KEYNOTE Invited speech: 14:30 - 15:30 SEQUOIA LECTURE ROOM Prof. Dr. Goran Pavić COFFEE BREAK 15:30 - 15:50 MAIN HALL AFTERNOON SESSIONS 15:50 - 17:30 SEQUOIA AND QUERCUS LECTURE ROOM CLOSING CEREMONY 17:30 - 18:00 SEQUOIA LECTURE ROOM 10 Program Registration WEDNESDAY, SEPTEMBER 20 8:00 - 9:00 Opening ceremony WEDNESDAY, SEPTEMBER 20 9:00 - 9:30 Assoc. Prof. Dr. Mateja Dovjak, Congress Chair, President of the Slovenian Acoustical Society (SDA), Faculty of Civil and Geodetic Engineering of the University of Ljubljana, Slovenia Welcome speech On behalf of Slovenian Acoustical Society and Faculty of Civil and Geodetic Engineering of the University of Ljubljana, Assoc. Prof. Dr. Mateja Dovjak, Congress Chair, President of the SDA, Slovenia On behalf of InnoRenew CoE, Prof. Dr. Andreja Kutnar, director of the InnoRenew CoE, Slovenia On behalf of Hungarian Scientific Society for Optics, Acoustics, Motion Picture and Theatre Technology (OPAKFI), Beáta Mesterházy, M.Sc., President of the Department of Acoustics of OPAKFI, BME (Budapest University of Tecnology and Economics), Department of Building Constructions, Laboratory of Building Acoustics On behalf of Hrvatsko akutsičko društvo, Acoustical Society of Croatia (HAD), Prof. Dr. Marko Horvat, President of the HAD, Faculty of Electrical Engineering and Computing, University of Zagreb, Croatia 11 WEDNESDAY, SEPTEMBER 20 - morning session SEQUOIA LECTURE ROOM QUERCUS LECTURE ROOM SS04 - ROOM ACOUSTICS SS01 - NOISE AND VIBRATIONS Session chair: Prof. Dr. Kristian Jambrošić, University of Zagreb Session chair: Prof. Dr. Antonio Petošić, University of Zagreb, Faculty of Electrical Engineering and Computing Faculty of electrical Engineering and Computing TIME TITLE PRESENTER TIME TITLE PRESENTER Kristian Jambrosic Integration of Acoustic Design of a (University of Zagreb Psychoacoustic Nejc Cerkovnik (University of New Concert Venue for 10:50 Faculty of Electrical 10:50 Perception for Ljubljana, Faculty of Classical Music in Split, Engineering and Enhanced Design of Mechanical Engineering) Croatia Computing) Axial Fans Noise generating mechanisms Real or Synthetic? A analysis and its Machine Learning Marko Pap (School of optimization on 11:10 Approach to Classifying Electrical Engineering, 11:10 Andrej Biček (Nela d.o.o) electronical Room Impulse University of Belgrade) commutated wet- Responses dry vacuum cleaner suction unit Modeling and LCA study of different assessment wind Antonio Petošić (University of recycled sound Urban Kavka (InnoRenew turbine noise at 11:30 11:30 Zagreb, Faculty of Electrical absorbers from CoE) different Engineering and Computing) melamine foam waste meteorological conditions Evaluation of model Multichannel based noise Reverberation Time Andrej Hvastja (University protection study Krešimir Burnać (Faculty of 11:50 Measurements of of Ljubljana, Faculty of 11:50 based on in-situ Civil Engineering, University Miura-Ori Origami in Mechanical Engineering) vibro-acoustic of Zagreb) Alpha Chamber railway track analysis Steel reilway bridge noise, lack of Reverberation time reduction effect on Andrea Andrijasevic estimation from airborne noise due 12:10 (Polytechnic of Rijeka, 12:10 Rok Rudolf (ZAG) emotional speech to Croatia) signals vibration dampers possibly acting a s noise sources Determination of the Psychoaoustics of position of equipment Pseudosound in noise sources in Mateja Dovjak (University Jurij Prezelj (University of Turbulent flow of 12:30 educational institutions of Ljubljana, Faculty of Civil 12:30 Ljubljana, Faculty of Centrifugal Fan used according to and Geodetic Engineering) Mechanical Engineering) in Household subjectively evaluated Appliance speech intelligibility 12 WEDNESDAY, SEPTEMBER 20 - afternoon session SEQUOIA LECTURE ROOM QUERCUS LECTURE ROOM SS02 - ADVANCED MEASUREMENT TECHNIQUES IN SS06 - ACOUSTIC SOFTWARE AND TRAINING ACOUSTICS Session chair: Assoc. Prof. Dr. Mateja Dovjak, University of Session chair: Assist. Prof. Dr. Rok Prislan, InnoRenew CoE Ljubljana, Faculty of Civil and Geodetic Engineering TIME TITLE PRESENTER TIME TITLE PRESENTER Audio exercises: Daniel Svenšek Challenges in the quality, pitch Andrea Andrijasevic (University of Ljubljana, 15:50 introduction of timbre 15:50 statistics and long- (Polytechnic of Rijeka, Faculty of Mathematics coordinates for violoncelli term spectra of the Croatia) and Physics) sound files Acoustics Knowledge Alliance project: the Soundscape monitoring development of Marko Horvat (University system for earthquake- Karlo Filipan (Catholic open-access of Zagreb, Faculty of 16:10 16:10 affected urban spaces – University of Croatia) interactive online Electrical Engineering and Zagreb case study educational Computing) materials in acoustics - strategy and results Immission Directivity as a Jurij Prezelj (University 16:30 tool for generation of of Ljubljana, Faculty of Noise Maps Mechanical Engineering) Data Selection for Reduced Training Effort in Stefan Grebien 16:50 Vandalism Sound Event (Joanneum Research) Detection Experimental sound field characterization with Rok Prislan (InnoRenew 17:10 automated high- CoE) resolution impulse response measurements 13 THURSDAY, SEPTEMBER 21 - morning session SEQUOIA LECTURE ROOM QUERCUS LECTURE ROOM SS05 - BUILDING ACOUSTICS SS01 - NOISE AND VIBRATIONS Session chair: Beáta Mesterházy, M.Sc., President of Session chair: Prof. Tino Bucak, HAD, University of Zagreb, the Department of Acoustics of OPAKFI, BME Faculty of Transport and Traffic Sciences, Croatia TIME TITLE PRESENTER TIME TITLE PRESENTER ILEGAL USE OF FIRECRACKERS Design and construction AND ITS CONSEQUENCES - of a temporary test CASE STUDY OF HUMAN Ferdinand Deželak 10:50 facility for sound Nika Šubic (MK3 d.o.o.) 10:50 RIGHTS VIOLATION AT (Retired researcher) insulation SLOVENIAN COURTS – PART I: measurements of doors LEGISLATION AND COURT PROCEEDINGS Acoustic performance of buildings, ILEGAL USE OF FIRECRACKERS components and AND ITS CONSEQUENCES - Franz Dolezal (IBO - materials as a CASE STUDY OF HUMAN Ferdinand Deželak 11:10 Austrian Institute for 11:10 parameter for RIGHTS VIOLATION AT (Retired researcher) Building and Ecology) ecological and social SLOVENIAN COURTS – PART sustainability II: PHYSICAL BACKGROUND assessments Possible applications of high-performance Samo Beguš floating floors and Low frequency noise (University of Zoltán Horváth (CDM 11:30 consequent relevant 11:30 measurement in the Ljubljana, Faculty of Stravitec Kft.) dimensions of passenger cabin Electrical performance and design Engineering) criteria Improvement of impact sound insulation with Beáta Mesterházy, Measurement and Egon Susič (Danfoss 11:50 tile underlay materials – (Department of Acoustics 11:50 Characterization of Control Trata d.o.o.) impact sound insulation of OPAKFI, BME) Valves Noise without floating floors SEQUOIA LECTURE ROOM SS03 - SOUNDSAPE AND SOUND REPRODUCTION TECHIQUES Session chair: Prof. Dr. Marko Horvat, President of the HAD, Faculty of Electrical Engineering and Computing, University of Zagreb TIME TITLE PRESENTER The soundscape of Alberto Quintana- university campuses: Gallardo (Centre for 12:10 sound essays on the Physics Technologies Polytechnic University of (CTFAMA), Universitat València Politècnica de València) Accuracy of Dynamic Sound Source Localization Vedran Planinec (Faculty in Binaural Audio Systems of Electrical Engineering 12:30 with Head-Tracking and Computing, Utilizing Generic and University of Zagreb) Individual HRTFs 14 THURSDAY, SEPTEMBER 21 - afternoon session SEQUOIA LECTURE ROOM QUERCUS LECTURE ROOM SS03 - SOUNDSCAPE AND SOUND REPRODUCTION TECHIQUES SS01 - NOISE AND VIBRATIONS Session chair: Prof. Dr. Marko Horvat, President of the HAD, Session chair: Dr. Egon Susič, Danfoss Trata d.o.o., Slovenia Faculty of Electrical Engineering and Computing, University of Zagreb TIME TITLE PRESENTER TIME TITLE PRESENTER Proposing noise barriers along Application of existing national Mihael Žiger (Nacionalni Ambisonics to Building 15:50 Armin WILFLING (IBO) 15:50 roads through laboratorij za zdravje, okolje Acoustics – Challenges settlements - the in hrano) and Opportunities case of Novo mesto and its surroundings A Comparison between Measurement and Real and Reproduced modelling Antonio Petošić (University of Martina Vrhovnik 16:10 Sound Fields for Impact 16:10 uncertainty in Zagreb, Faculty of electrical (InnoRenew CoE) Noise Annoyance accredited acoustic Engineering and Computing) Ratings procedures Development of a special standard for outdoor music Luka Čurović (University of 16:30 events in Slovenia Ljubljana, Faculty of based on Mechanical Engineering) measurements and calculations SEQUOIA LECTURE ROOM SS02 - ADVANCED MEASUREMENT TECHNIQUES IN ACOUSTICS Session chair: Assist. Prof. Dr. Samo Beguš, University of Ljubljana, Faculty of Electrical Engineering TIME TITLE PRESENTER Estimating Speed of Sound in Granular Materials: Impulse Anže Železnik (University 16:30 Response Extraction and of Ljubljana, Faculty of Wave Decomposition in Mechanical Engineering) an Extended Impedance Tube A smart method to Sharath Peethambaran calibrate universal Subadra (Hochschule für 16:50 testing machines by Angewandte incorporating acoustic Wissenschaften Hamburg) methods Unsupervised Classification of Welding Jurij Prezelj (University of 17:10 processes based on Ljubljana, Faculty of Psychoacoustic Sound Mechanical Engineering) Features 15 Closing ceremony THURSDAY, SEPTEMBER 21 17:30 - 18:00 Assoc. Prof. Dr. Mateja Dovjak, Congress Chair, President of the Slovenian Acoustical Society (SDA), Faculty of Civil and Geodetic Engineering of the University of Ljubljana, Slovenia An invitation to the 11th AAAA in Zagreb 2025: Prof. Dr. Marko Horvat, President of the HAD, Faculty of Electrical Engineering and Computing, University of Zagreb, Croatia 16 Keynote Invited speech High-speed camera based identification of sound and vibrations Prof. Dr. Janko Slavič Ladisk, Faculty of Mechanical Engineering, University of Ljubljana E-mail: janko.slavic@fs.uni-lj.si Professor Dr. Janko Slavič is a full professor at the Faculty of Mechanical Engineering at the University of Ljubljana. Prof. Slavič is a recipient of the Fulbright scholarship (University of Texas at Austin 2005-2006) and is a co-author of 82 scientific articles, 56 of which are in category Q1. His works have been cited over ten thousand times. As a mentor or co-mentor, he has supervised 16 completed doctoral theses. In the last 10 years, his research has made significant contributions to the development of science in four scientific areas: vibration fatigue, experimental modal analysis based on high-speed camera recordings, 3D printing of sensors, and research based on open-source code. Image-based measurement techniques have recently gained popularity and are increasingly used in various applications as a viable alternative to traditional measurement methods. Image-based techniques offer high spatial density of information, and with advancements in hardware, they can also provide a high frequency of image acquisition, such as 20k frames per second at megapixel resolution. In his lecture, dr. Slavič will review the origins of image-based methods and explain the well-established classical method of digital image correlation (DIC). Advanced signal processing methods will be presented, as well as the challenges associated with noise and overdetermination in image data. It will be shown that amplitude identification is successful for harmonic motion up to the resolution of 1/100,000th of a pixel. Finally, selected methods will be explored that have great potential for future research, such as 3D vibration reconstruction based on frequency domain triangulation and the spectral optical flow imaging experimental technique. 17 Contributed papers Room acoustics 1. Acoustic Design of a New Concert Venue for Classical Music in Split, Croatia Kristian Jambrosic (University of Zagreb Faculty of Electrical Engineering and Computing) 2. Real or Synthetic? A Machine Learning Approach to Classifying Room Impulse Responses Marko Pap (School of Electrical Engineering, University of Belgrade 3. LCA study of different recycled sound absorbers from melamine foam waste Urban Kavka (InnoRenew CoE) 4. Multichannel Reverberation Time Measurements of Miura-Ori Origami in Alpha Chamber Andrej Hvastja (University of Ljubljana, Faculty of Mechanical Engineering) 5. Reverberation time estimation from emotional speech signals Andrea Andrijasevic (Polytechnic of Rijeka, Croatia) 6. Determination of the position of equipment noise sources in educational institutions according to subjectively evaluated speech intelligibility Mateja Dovjak (University of Ljubljana, Faculty of Civil and Geodetic Engineering) 18 ACOUSTIC DESIGN OF A NEW CONCERT VENUE FOR CLASSICAL MUSIC IN SPLIT, CROATIA Kristian Jambrošić 1, Marko Horvat 2 University of Zagreb, Faculty of Electrical Engineering and Computing, Unska 3, Zagreb, Croatia 1 kristian.jambrosic@fer.hr, 2 marko.horvat@fer.hr Abstract: This paper presents the key points of acoustic design of a new concert venue in Split, based on the revitalization of an old building. This venue is dedicated to showcasing classical music performances by soloists and small orchestras. It is built as a classical shoebox-shaped main hall with a small balcony and a seating capacity of 250 that is used for small ensemble performances, accompanied by a smaller hall accommodating up to 60 attendees that is used for mostly solo performances. Throughout the design and construction phases, careful consideration was given to preservation of the historical and architectural value of the building. The acoustic design process aimed to create an exemplary venue lauded for its superior sound quality by both the audience and performers. The paper outlines the decision-making process behind the selection of appropriate interior materials and the placement of acoustic elements, to achieve optimal acoustics in the halls. Furthermore, the paper presents the results of comprehensive acoustic measurements conducted during the construction period, highlighting the effectiveness of the implemented design. By evaluating the acoustical performance of the halls before and after their refurbishment, this study provides valuable insights into the successful integration of architectural design and acoustic considerations. Keywords: concert hall acoustics, stage design, room acoustics simulation, acoustic measurements 1. INTRODUCTION strong Art Nouveau expression, became an imposed approach to the renovation, Fig. 1. The representative building of the Croatian Home With the establishment of the company “Public Institution (“Hrvatski dom” in Croatian) in Split was constructed in in Culture, Croatian Home Split”, the town of Split gained 1908 as the headquarters of Croatian nationalist societies, a professional concert hall for the first time in its history. according to the design by engineer Kamilo Tončić. The project was created in accordance with the contemporary tendencies of the Viennese Wagner School. Art Nouveau influences can be seen in the design of the main facade, enriched with geometric and stylized floral decorations, as well as in the interior of the Ceremonial Hall on the upper floor. But, the hall was completely neglected from the early years after World War II. Preparations for the renovation of the building began in 2005 with a clear idea of its future purpose, which would include a puppet theatre on the ground floor and a multipurpose concert hall on the upper floor, with all other spaces serving the purpose of concert activities. The comprehensive restoration of the physically destroyed original appearance of Tončić's work, characterized by a Fig. 1. Renovated facade of the Croatian Home building. Jambrošić et al.: ACOUSTIC DESIGN OF A NEW CONCERT VENUE FOR CLASSICAL MUSIC IN SPLIT, CROATIA 19 AAAA – 2023 – IZOLA - Conference Proceedings The Ceremonial Hall, reconstructed in detail as it looked like before the World Was II, has a seating capacity of 250 visitors, 216 of which are seated on the ground floor and 34 on the balcony in the back of the hall, Fig. 2 and 3. Fig. 4. The interior of the Jakov Gotovac Music Salon, view towards its back. Fig. 2. The interior of the Ceremonial Hall, view from the beginning of the stage towards its back. Fig. 5. The interior of the Jakov Gotovac Music Salon, view towards its stage. Fig. 3. The interior of the Ceremonial Hall, view from the balcony towards its stage. 2. ACOUSTIC DESIGN OF THE CONCERT VENUES The smaller hall, the Jakov Gotovac Music Salon, did not exist as a public event space before. A large room in the The task of the authors of this paper was to provide to the topmost floor has been adapted as a multipurpose space venue owners an acoustic design project for both concert because of high demands for such smaller venues. It can halls prior to the reconstruction work. It had to predict the accommodate between 50 and 70 visitors, Fig. 4 and 5. acoustic quality of both halls and to prove that they will Both spaces are equipped with top-quality audio-visual be appropriate for their intended purposes. The equipment, allowing for a high production level of musical Ceremonial Hall is primarily built for concerts of classical and other events. In addition to musical and cultural music by small orchestras and soloists, and for occasional purposes, the spaces are suitable for various events such use as a venue for speech, choirs, jazz music, and other as congresses, lectures, presentations, anniversary events that will fit the owner’s need. The Music Salon, celebrations, and more. given its size, primarily aims to host speech events (conferences, drama plays), but also soloist concerts. Jambrošić et al.: ACOUSTIC DESIGN OF A NEW CONCERT VENUE FOR CLASSICAL MUSIC IN SPLIT, CROATIA 20 AAAA – 2023 – IZOLA - Conference Proceedings To simulate the acoustics of both halls, 3D wireframe on the type of music performed (organ music, symphony models were prepared, and detailed simulation have been orchestra, chamber music, solo performances and done in the Odeon Auditorium room acoustic simulation recitals, dance music, amplified modern music). The next software [1]. Wireframe models for both halls can be seen important feature of the hall that determines the optimal in Fig. 6 and Fig. 7. range of acoustic parameters is its size, that is, the volume of the hall. All of this should be taken into account when assessing and evaluating the acoustic quality of halls, as stated in the reference standards and publications that describe and determine the acoustic properties of the space. The requirements from the EN ISO 3382-1:2009 standard cover the widest range of uses, which is why they are the least strict ones [2]. A finer distinction is given by the criteria according to Beranek [3] and Rossing [4], with the fact that Beranek covers more parameters and purposes, and Rossing deals with halls whose volume is outside Beranek's criteria. The rating according to Long [5] is used to determine the specific volume (m3/seat) and acoustic conditions for speech purposes. A good reference is also Fig. 6. 3D wireframe model of the Ceremonial Hall. the book by Barron [6]. 2.1. The Ceremonial Hall Z O X Y The volume of the hall is about 1,650 m3, which places it the category of smaller halls intended primarily for chamber music performances. From the volume of the hall and the number of seats, the specific volume is obtained, that is, the volume per seat, which in this case is 6.6 m3/seat. The shape of the hall is particularly interesting since it is Odeon©1985-2021 Licensed to: University of Zagreb, Croatia built as a shoe-box type hall with the dimensions almost ideally half the length, width, and height of the world most Fig. 7. 3D wireframe model of the Jakov Gotovac Music famous hall for classical orchestra music, the Musikverein Salon. Hall in Vienna. Only the height of the Ceremonial Hall is around one meter smaller than half the height of the Measurements of the room acoustic parameters were Vienna Hall. Thus, the shape of the hall is already a good planned as part of the acoustic design project after the indication that that, at least from the perspective of its reconstruction works to prove the quality of the acoustic geometry, it should function acoustically very well! design. It must be noted that there was little space for adding any The criteria for the acoustic design and the optimal values acoustic elements that would diverge the occurrence of of acoustic parameters depend on several important the Ceremonial Hall from its original looks as this was the features of the halls. The first is certainly the purpose of prerequisite of the conservators. The only possibility was the halls, where one should distinguish between halls to carefully chose the materials of the surfaces and intended for musical performances (concert halls of all especially audience seats, with a little more freedom in sizes), halls intended for speech (theatres, cinemas, the stage ceiling since it was not visible form the audience multifunctional halls), as well as halls intended for both due to an existing portal (Fig. 3). Therefore, the music and speech (opera, theatres). Furthermore, the reverberation time of the hall was tuned with a proper acoustic parameters of halls intended for music depend amount of the porous acoustic absorbers mounted on the Jambrošić et al.: ACOUSTIC DESIGN OF A NEW CONCERT VENUE FOR CLASSICAL MUSIC IN SPLIT, CROATIA 21 AAAA – 2023 – IZOLA - Conference Proceedings ceiling above the stage where stage technique equipment small volume, the concert chairs significantly affect the is mounted, but not necessarily in enough extent to acoustics of the room, it was necessary to predict very prevent flutter echo for the musicians. precisely the sound-absorbing characteristics of the Based on the above considerations, the value ranges of concert seats. Therefore, in consultation with the owners, the basic acoustic parameters, i.e. reverberation time RT60 subsequent measurements of the acoustic parameters in and clarity of music C80, were chosen for primarily music the salon were carried out to measure the impact of the performances, according to [2] and [7]. finally selected concert chairs. In particular, the value of the reverberation time at According to [7], the specific volume per seat for a room medium frequencies that needs to be achieved in a full for speech and music events should be in the range hall should be in the range of 1.5 to 1.7 seconds. In an between 6 and 8 m3/seat, and for exclusively musical empty hall, an increase in the reverberation time of up to events it should be greater than 7 m3/seat. Considering 10%, to a maximum of 1.85 seconds, is permissible. The the volume of the hall of 233 m3, to meet the above specified range represents the optimal values for the conditions, the hall should have a maximum of 39 seats Ceremonial Hall, considering its size and intended for the audience for mixed speech-music performances, purpose. that is, no more than 33 seats for exclusively musical The average value of the clarity of music at medium performances. The planned 50 - 60 seats for the audience frequencies for the entire auditorium should ideally range exceed the stated figures, so it is necessary to keep in from -2 to +3 dB. mind that an excessive number of visitors leads to The range of possible values of the early sound decay time excessive acoustic damping of the space, which directly is not defined because the initial decay of the sound level affects the quality and experience of the musical in the room is defined primarily by the layout of the performance. acoustically treated surfaces. In this case, it is fixed and For primary musical performances, according to [7] the cannot be changed, which is why the amount of early recommended reverberation time RT60 is 1.1 seconds, decay time cannot be influenced. given as an average value in tertiary bands with central frequencies from 200 to 2500 Hz. 2.2. The Jakov Gotovac Music Salon There were no constrains by the conservators regarding 3. ACOUSTIC MEASUREMENTS OF THE CONCERT VENUES the acoustic design of the Music Salon, in terms of the amount and placement of acoustic elements in the room. The measurements of acoustic parameters were carried The proposed solution for the acoustic treatment of the out to determine the acoustic situation in the hall after salon must solve the problems of the parallel side walls on restoration with installed acoustic treatment. the stage and the front and back walls of the hall, which The measurement of the acoustic parameters of the stage are also parallel. of the Ceremonial Hall was also carried out in a later stage The solution consists of placement of medium- of the projects in order to evaluate the acoustic situation upholstered concert chairs in the auditorium, the sound in the area of the stage. absorption of which largely determines the overall All measurements were carried out by the impulse reverberation in the room and enables the relative response integration method using a sinusoidal signal of uniformity of the hall's reverberation, regardless of the continuously variable frequency as an excitation signal. number of listeners present, and installation of sound Impulse responses of the spaces were obtained from diffusers on the side walls of the stage, and additionally which it is possible to calculate the values of all relevant diffusers on the wall opposite the stage, at the entrance acoustic parameters depending on the frequency with the to the salon. corresponding values of the relevant single parameters. Typical value ranges of individual acoustic parameters At the time of measurement, the halls were empty, i.e. defined in [2] for concert and multi-purpose halls show without an audience, but fully equipped and completed that the early decay time EDT is typically in the range of 1 according to the renovation projects and in accordance to 3 seconds, while the clarity of music C80 is typically in with the acoustics project. the range of -5 to 5 dB. Given that in a room of such a Jambrošić et al.: ACOUSTIC DESIGN OF A NEW CONCERT VENUE FOR CLASSICAL MUSIC IN SPLIT, CROATIA 22 AAAA – 2023 – IZOLA - Conference Proceedings 3.1. Ceremonial Hall 3.2. Jakov Gotovac Music Salon Fig. 8. shows the averaged reverberation time over all Fig. 10. shows the averaged reverberation time over all measured locations in the whole frequency range of measured locations in the whole frequency range of interest (from 50 Hz to 10,000 Hz in 1/3 octave bands), interest (from 50 Hz to 10,000 Hz in 1/3 octave bands). and Fig. 9 the averaged clarity C80 in the same frequency range. Fig. 10. Average reverberation time vs frequency in the Jakov Gotovac Music Salon. Fig. 8. Average reverberation time vs frequency in the Ceremonial Hall. It is evident that the reverberation time in Jakov Gotovac's salon is at the lower limit of tolerance, even when it is empty. This is the result of sound absorption due to the large number of upholstered concert chairs. The short reverberation time at low frequencies is more suitable for speech than for music and is a direct consequence of the presence of a large amount of plasterboard in the suspended ceiling. Whenever the program allows it, it is recommended to reduce the number of chairs for visitors to or below the maximum recommended number, which will achieve greater liveliness of the space in terms of acoustics. This will have a favourable effect on the experience of the performed music, both for the visitors and the Fig. 9. Average clarity of music vs frequency in the performers. Ceremonial Hall. The single-number values for both parameters are: 4. STAGE ACOUSTICS OF THE CEREMONIAL HALL Reverberation time RT60, 400-1250 Hz = 1.57 s During the initial acoustic project of the Ceremonial Hall Clarity C80, 400-1250 Hz = 0.05 dB in 2015, an important limitation in terms of possible intervention in the interior of the hall as part of the Bot values fit perfectly the expected optimum values for acoustic treatment project was the conservator's request this hall, thus proving that, at least from the acoustic that the hall be completely renovated in accordance with parameters standpoint of view, the hall might be the original plans from the first half of the 20th century. considered to be successful. Accordingly, the acoustic treatment of the hall had to be designed to fully respect its original form. For this reason, Jambrošić et al.: ACOUSTIC DESIGN OF A NEW CONCERT VENUE FOR CLASSICAL MUSIC IN SPLIT, CROATIA 23 AAAA – 2023 – IZOLA - Conference Proceedings it was not possible to change the shape of the stage or add Another solution is to bevel the parallel side walls of the any additional acoustic elements to the walls of the hall or stage at an angle greater than 7°, as shown in Fig. 12. This stage. solution prevents the occurrence of a fluttering echo but The concert hall has a built-in portal opening between the does not provide an additional reflective surface that stage and the rest of the hall. The upper part of the portal would target sound energy from the stage to the opening (overhang), unfortunately, prevents the use of auditorium area, as is the case with the orchestra shell. ceiling surfaces as natural reflectors of sound towards the The owners did not accept this solution either, because it auditorium, both those above the stage itself and those would reduce the area of the stage, which is too small for directly in front of the portal. some performances as it is. Moreover, soon the owners intend to replace the existing stage with a new, sector- 4.1. Proposed solutions variable stage with a system of height-adjustable modules. Based on the conversation with the owners and through past experiences of musical performances in the concert hall since its opening, the following problems with the acoustics of the stage were recognized: 1. The parallel smooth side walls of the stage cause a fluttering echo when any transient sounds occur, which bothers the performers on the stage. 2. The space of the stage, which is bounded by the side parallel walls, the back smooth wall, and the opening of the portal, is subjectively too loud for the Fig. 12. Second version of the stage acoustics solution. performers, especially when there is an ensemble on the stage with many performers who then try to Finally, the only possible solution to the stage acoustic reduce the volume of their performance. problems is to install diffuser elements on the existing 3. The stage, as well as the hall, is too small for large walls. The basic sketch of the solution is shown in Fig. 13., ensembles, although there is a need to hold musical and the appearance of the stage is shown in Fig. 14. events that precisely include such ensembles. The optimal solution to these problems is the use of an orchestra shell (fixed or movable). It eliminates parallel surfaces and the fluttering echo associated with them, and in lowers the problem of excessive volume by scattering the sound energy in all directions. The basic sketch of this solution is shown in Fig. 11. The owners have confirmed on several occasions that this solution is not acceptable due to the lack of space for storing elements of the orchestra shell, i.e. the lack of a Fig. 13. Third version of the stage acoustics solution. service room in the immediate vicinity of the stage. Fig. 11. Fist version of the stage acoustics solution. Fig. 14. The stage after installation of diffuser elements. Jambrošić et al.: ACOUSTIC DESIGN OF A NEW CONCERT VENUE FOR CLASSICAL MUSIC IN SPLIT, CROATIA 24 AAAA – 2023 – IZOLA - Conference Proceedings 4.2. Influence of the diffusing elements The acoustic parameters of the stage are defined to evaluate its acoustic properties from the perspective of musicians, or performers in general. At the same time, it is important that the acoustic conditions in the hall enable the performers on stage to hear each other well and to hear well the response of the hall itself. According to [2], the STEarly parameter shows how well the musicians on the stage can hear each other, while the STLate parameter quantifies the response of the hall itself from the performer's position, i.e. its perceived resonance. Both parameters are obtained in a similar way as the acoustic parameters of the hall, i.e. from measured impulse responses at certain points on the stage. Fig. 16. Average clarity of music on the stage vs The single value of the ST frequency after the installation of diffusers. Early and STLate parameter in the middle frequency range for all measurement points, i.e. the average value for the entire stage, before the The shortening of the EDT early decay time and the installation of the diffusers, are: increase in the C80 music clarity values indicate an increase in the energy of direct sound compared to the energy of ST reverberant sound in the stage space. In other words, the Early, 250-2000 Hz = -8.5 dB ST performers of musical (and spoken) performances will Late, 250-2000 Hz = -11.9 dB hear their own performance and the performance of The values of the same parameters after the installation other members of the ensemble more clearly after the of the diffusers are: intervention and, consequently, will more easily perform phrases with a faster tempo and greater dynamics, ST compared to the situation before adding the diffuser. Early, 250-2000 Hz = -8.9 dB ST The global change in the acoustic conditions on the stage Late, 250-2000 Hz = -12.9 dB is also reflected in the difference in the single-digit The installation of diffusers lowered the ST average values of the ST Early and STLate Early and STLate parameters values. Moreover, this had also some influence on measured before and after the mentioned intervention. lowering the reverberation time of the hall and raising the In particular, the values of the STEarly parameter obtained clarity of music, as can be seen in Fig. 15. and Fig. 16. by measuring on the stage of the hall in question before the intervention amount to -8.5 dB for all measurement points. After the intervention, the values of the specified parameter measured and calculated in the same way drop to -8.9 dB for both methods of calculation. Since this parameter basically expresses the ratio of the total energy of early reflections arriving from the stage space itself and the energy of direct sound, the obtained values indicate that the sound volume on the stage itself will be somewhat lower after the acoustic intervention, but not significantly. A significant reduction of the sound volume on the stage is possible only a) by adding a large amount of extremely sound-absorbing material, which would significantly impair the acoustic properties of the hall, Fig. 15. Average reverberation time vs frequency in the and/or b) by limiting the number of performers on the Ceremonial Hall after the installation of diffusers. Jambrošić et al.: ACOUSTIC DESIGN OF A NEW CONCERT VENUE FOR CLASSICAL MUSIC IN SPLIT, CROATIA 25 AAAA – 2023 – IZOLA - Conference Proceedings stage, which would set limits on the variety of the musical exceeds their optimal number and thus add to much program performed in the hall. absorption to the room. The values of the STLate parameter obtained by the However, to receive a final confirmation of the acoustical measurement before the intervention amount to -11.9 dB quality of both venues, it was necessary to wait for the for all measurement points. After the intervention, these verification of real audience visiting actual music values drop to -12.9 dB. This result indicates that the performances. In general, the audience reported that the response of the hall from the listening point of the acoustic comfort in the halls is very good, and that the musicians themselves will be slightly weaker than before overall clarity of music performances is high, even on the the intervention, but still very strong. balcony of the Ceremonial Hall. Some of the performers in The values of both parameters measured both before and some orchestras had objections to the sound on stage, after the intervention are at the upper limit of the typical especially the high volume. By subsequently installing the range of values and indicate that the musicians on stage diffusers on the stage walls, this problem was also can hear each other very well both with and without successfully solved. The sound of the performance did not installed diffusers. change significantly in the audience, and it was confirmed The influence of the installed diffusers can be seen very that the volume on the stage was lower, with a reduced clearly in the impulse responses measured on the stage of probability of flutter echo occurrence. the hall. Fig. 17. shows an example of two impulse responses measured at the same point on the stage before and after the installation of the diffuser. It is clear 6. REFERENCES that the sound energy contained in strong individual reflections from smooth wall surfaces is truly dispersed [1.] Odeon Auditorium 17 Room Acoustic Software. and divided into much weaker components, which was Available at: https://odeon.dk/ (Accessed 28 August 2023). the goal of installing the diffuser. [2.] ISO 3382-1:2009. Acoustics — Measurement of room acoustic parameters — Part 1: Performance spaces. [3.] Beranek, L. Concert halls and opera houses: music, acoustics and architecture. Springer-Verlag, 2004. [4.] Rossing, D. Springer Handbook of Acoustics. Springer Verlag, 2007. [5.] Long, M. Architectural Acoustics. Academic Press, 2006. [6.] Barron, M. Auditorium Acoustics and Architectural Design, Sp on Press, 2009. [7.] DIN 18041:2012. Audibility of rooms. Fig. 17. Impulse response measured in the same location of the stage, before (left) and after (right) adding diffusers. 5. CONCLUSION The process of the acoustical design of the two halls of the Croatian House in Split, the Ceremonial Hall, and the Jakov Gotovac Salon, proved to be very successful. The correct modelling of both halls and the subsequent acoustic measurements are very well in agreement. Moreover, the measured acoustical parameters are in their optimal ranges. The only deviation can be seen in the Jakov Gotovac Salon when the number of audience chairs Jambrošić et al.: ACOUSTIC DESIGN OF A NEW CONCERT VENUE FOR CLASSICAL MUSIC IN SPLIT, CROATIA 26 REAL OR SYNTHETIC? A MACHINE LEARNING APPROACH TO CLASSIFYING ROOM IMPULSE RESPONSES Marko Pap1, Miloš Bjelić2 1 School of Electrical Engineering, University of Belgrade, Belgrade, Serbia 2 School of Electrical Engineering, University of Belgrade, Belgrade, Serbia Abstract: machine learning. Nonetheless, there remains a need for a clas- This paper presents an innovative dual-branch neural network sifier that is more efficient, accurate, and comprehensive, espe- classifier designed to discern between real and synthetic room cially when handling diverse data types. impulse responses. Responding to the critical need for verify- To address this challenge, we have designed an innovative ing the authenticity of room impulse responses used in applicadual-branch neural network classifier. Our approach integrates tions such as audio forensics, acoustic environment modeling, the strengths of two potent neural network models - a Convo- and virtual reality, our classifier uniquely combines a Convolu-lutional Neural Network (CNN) and a Fully Connected Network tional Neural Network (CNN) and a Fully Connected Network (FCN). The CNN branch concentrates on processing spectrogram (FCN) to analyze both spectrogram data and a set of acousti- representations, thereby capturing time-frequency characteris- cal features. The CNN branch processes spectrogram represen- tics, while the FCN branch focuses on a set of specific acoustical tations, capturing time-frequency characteristics, while the FCN features [3]. Recent advancements in neural-network-based RIR branch attends to a set of predefined acoustical features, en- generation and analysis, such as the FAST-RIR [4] and deep prior abling comprehensive analysis. This dual-branch approach al- approaches [5], highlight the potential of such methods in the lows the classifier to leverage the strengths of both networks, field. contributing to an exceptional performance of 99% accuracy on The acoustical features extracted for the FCN include: the test set. We provide a detailed description of the classifier’s • Spectral centroid architecture, underlining the features and design choices that enable its high accuracy. Our results imply the classifier’s po- • Spectral bandwidth tential as a tool in maintaining the integrity and authenticity of • Spectral flatness room impulse responses. We discuss the potential applications and directions for future work, adding a valuable contribution to • Spectral contrast the field of audio and acoustic research. • Zero-crossing rate Keywords: Dual-branch neural network, Room impulse re- • Root mean square (RMS) value sponses, Authenticity verification, Audio forensics, Acoustic en- • Mel-frequency cepstral coefficients (MFCCs) vironment modeling, Convolutional Neural Network (CNN), Fully Connected Network (FCN), Spectrogram data, Acoustical fea- • Energy split between early and late reflections of the tures, Classifier performance, Classifier architecture sound [6] These features are fundamental in providing a holistic un- 1. INTRODUCTION derstanding of RIRs, as demonstrated by recent works on room acoustical parameter estimation using deep neural networks [7] Room impulse responses (RIRs) are crucial in various applica- and virtually supervised learning for mean absorption estima- tions such as audio forensics, acoustic environment modeling, tion [8]. and virtual reality, due to the intricate acoustical properties they Our dual-branch approach is designed to harness the capture [1,2]. Determining whether these responses are authen- strengths of both the CNN and the FCN, thereby enhancing the tic or synthetically generated is vital for maintaining the integrity classifier’s overall performance. With an accuracy rate of 99% of data utilized within these domains [1]. on the test set, our classifier demonstrates high effectiveness. Identifying the genuineness of RIRs is a complex task that This paper provides a detailed exploration of our dual- has seen advancements through digital signal processing and branch neural network classifier’s architecture, emphasizing the Pap et al.: Real or Synthetic? A Machine Learning Approach to Classifying Room Impulse Responses 27 design choices and the extracted acoustical features that con- of nodes across layers (32, 64, and 128) aids this extrac- tribute to its high accuracy. We believe our classifier has the potion process, a design inspired by architectures like VG- tential to serve as a reliable tool in verifying the authenticity of GNet [11]. room impulse responses. Additionally, we also discuss potential 3. 3x3 Filters: Empirical research, such as the work by the applications and explore avenues for future research, aiming to Visual Geometry Group [11], supports the efficiency of provide a significant contribution to the audio and acoustic re- 3x3 filters, which are adept at recognizing local patterns. search field. 4. MaxPooling2D for Dimensionality Reduction: Following each convolutional layer, MaxPooling2D reduces the spa- 2. METHODS tial size of the feature maps, accelerating computations and inducing translational invariance. 2.1 Neural Network Model Architecture and Training 5. Flattening for FCN Integration: Prior to merging with the Our dual-branch neural network classifier is built using the Ten-FCN, the CNN’s two-dimensional output is flattened, fa- sorFlow framework, a popular choice for implementing neural cilitating its integration with fully connected layers. networks in various applications, including medical imaging [9]. The classifier architecture combines a fully connected network The outputs of both FCN and CNN branches are concate- (FCN) for acoustic feature analysis and a convolutional neural nated, followed by a Dropout layer for additional regularization. network (CNN) for spectrogram analysis, similar to the approach The final output layer has nodes corresponding to the number used in distributed acoustic sensors [10]. of classes (real and synthetic in this case), with a softmax activa-The FCN is particularly suited for the analysis of a variety of tion for probability distribution. acoustic features. The reasons for its design are as follows: Training the model entails the use of the Adam optimizer, a learning rate of 0.01, and the categorical cross-entropy loss func-1. Acoustic Feature Input: Using acoustic features of the tion. We’ve adopted early stopping (with 5 epochs patience) room impulse response (RIR) as the input provides a com- and restore the best weights when training ends. If the valida- pact representation of the RIR’s characteristics. By focus- tion loss doesn’t improve for 2 consecutive epochs, we employ ing on these features, the network can efficiently learn a learning rate reduction strategy, decreasing it by a factor of 0.5 patterns that are most relevant for the classification task. until reaching a minimum delta of 0.01. 2. Sequential Dense Layers: Two Dense layers sequentially process and transform the input features into higher- 2.2 Data Extraction and Suitability for RIR Classification level representations. The choice of 256 nodes in the first Data plays an indispensable role in training any machine learning layer allows for the extraction of a broad set of features, model. For our RIR classification task, we focus on using audio while the subsequent reduction to 64 nodes ensures a data files, from which we extract a range of spectral and tempo- focus on the most critical features. ral features using the Librosa library. 3. ReLU Activation: The ReLU activation function introduces We start by loading the audio file with a sample rate of non-linearity, allowing the network to capture complex 16,000 Hz, ensuring it is in mono and truncated or padded to relationships in the data. Additionally, it offers compu- a duration of 2.5 seconds. The decision to limit the audio to tational efficiency and mitigates the vanishing gradient 2.5 seconds stems from the nature of room impulse responses. problem. In many environments, especially those of interest in our study, the majority of critical acoustic information is contained within 4. Dropout for Regularization: After each Dense layer, the initial 2.5 seconds of the RIR. This includes the direct sound, Dropout is introduced to prevent overfitting. By ran-early reflections, and the start of the reverberation tail. By trun-domly deactivating certain neurons during training, it en- cating or padding the signals to this duration, we focus on this hances the model’s generalizability. rich, informative segment, ensuring that the model is exposed For spectrogram representations, the CNN is specifically de- to the most characteristic parts of the RIR while eliminating po- signed: tential noise or less informative parts that might occur in longer recordings. 1. Spectrogram Input: Given that spectrograms provide a This uniformity in data presentation further allows the two-dimensional representation of a signal’s frequency model to focus on salient feature patterns, promoting a consis- and time attributes, the CNN, with its ability to recognize tent training process and avoiding biases or inconsistencies that local spatial patterns, becomes apt for such inputs. might arise from variable data sizes or formats. 2. Hierarchical Convolutional Layers: Three convolutional layers ensure the extraction of hierarchical features. Ba- 2.2.1 Spectral Features sic patterns are captured in initial layers, while deeper The spectral features extracted include the spectral centroid, layers understand complex ones. The increasing number spectral bandwidth, spectral flatness, and spectral contrast. Pap et al.: Real or Synthetic? A Machine Learning Approach to Classifying Room Impulse Responses 28 These features provide insight into the tonal characteristics of and N is the total number of samples in the signal. the sound. The Root Mean Square (RMS) is a measure that provides the The spectral centroid is a measure that indicates where the square root of the average of the squares of the values in the center of mass of the spectrum is located. Mathematically, it is signal. Mathematically, it is defined as: defined as: v u N −1 PN −1 1 f (n) · X(n) u X R = t x(n)2 (6) C = n=0 (1) N PN −1 X(n) n=0 n=0 where f(n) is the frequency at bin n, X(n) is the magni- where x(n) is the nth sample in the signal, and N is the tude of the frequency bin n, and N is the total number of fre- total number of samples in the signal. quency bins. The spectral bandwidth is a descriptor that provides the 2.2.3 Mel Frequency Cepstral Coefficients (MFCC) width of the spectral band. It shows how wide the range of fre- Mel Frequency Cepstral Coefficients (MFCCs) are widely used for quencies in which most of the spectral content is located. Math- speech and audio analysis. The procedure to compute MFCCs ematically, it is defined as: involves multiple steps: The first step often involves pre-emphasizing the signal s PN−1(f(n) − C)2 · X(n) B = n=0 (2) to increase the amplitude of high-frequency bands. The pre- PN −1 X(n) emphasized signal y(n) is computed from the original signal n=0 where C is the spectral centroid calculated using Equation x(n) as: (1). Spectralflatnessmeasureshownoise-likeasignalis,asop- y(n) = x(n) − αx(n − 1) (7) posed to being tonal. A higher spectral flatness indicates a more where α is the pre-emphasis coefficient, typically around noise-like signal. It is defined as the ratio of the geometric mean 0.97. to the arithmetic mean of the magnitude spectrum: The pre-emphasized signal is divided into overlapping frames and each frame is windowed. A commonly used window 1/N QN −1 X(n) is the Hamming window: n=0 F = (3) 1 PN −1 X(n) N n=0 2πn Spectral contrast is defined for each spectral band as the dif- w(n) = 0.54 − 0.46 cos (8) N − 1 ference between its peak and valley. For each band k: The Fourier Transform is computed for each windowed frame to obtain its frequency spectrum. Ck = max X(f ) − min X(f ) (4) f ∈B A Mel filterbank, consisting of overlapping triangular filters, k f ∈Bk where B is applied to the power spectrum to extract frequency bands. k is the set of frequency bins in band k and X (f ) is the magnitude of the frequency bin corresponding to frequency The Mel scale aims to mimic the non-linear human ear percep- f . tion of pitch. Finally, the Discrete Cosine Transform (DCT) is applied to the 2.2.2 Temporal Characteristics log energies of the Mel frequencies to obtain the MFCCs: We also calculate the zero-crossing rate and root mean square N −1 (RMS) value, which capture the signal’s temporal characteristics. MFCC X πk(2n + 1) log(E (9) The Zero Crossing Rate is a measure that indicates the rate at k = n) cos 2N n=0 which the signal changes from positive to negative or vice versa. Mathematically, it is defined as: where En is the energy in the nth Mel filter, and N is the total number of Mel filters. This results in a set of coefficients that serve as the feature N −1 1 vector for the frame. The procedure is repeated for each frame X Z = |sgn(x(n)) − sgn(x(n − 1))| (5) N − 1 to obtain a feature matrix for the entire signal. n=1 where sgn(x) is the sign function, defined as: 2.2.4 Energy Distribution of Early and Late Reflections For Room Impulse Response (RIR) classification, the energy dis- 1 if x > 0  tribution between the early reflections (first 5% of the audio file)  sgn(x) = 0 if x = 0 and the late reflections (remaining 95%) can provide crucial in-  −1 if x < 0 formation. Pap et al.: Real or Synthetic? A Machine Learning Approach to Classifying Room Impulse Responses 29 To compute these features, the signal x(n) is divided into These diverse and rich features ensure our dataset’s aptness two parts: xearly(n) and xlate(n). The energies Eearly and Elate for RIR classification. The amalgamation of spectral, temporal, for these segments are calculated as: and cepstral features encapsulates both local and global charac- teristics of the room impulse responses, permitting an exhaus- Nearly−1 tive analysis and thus promising high-accuracy classification. X Eearly = xearly(n)2 (10) n=0 2.3 Dataset Description Nlate−1 Our dataset was categorized into two primary divisions: Real X Elate = xlate(n)2 (11) Room Impulse Responses (RIRs) and Synthesized Impulse Re- n=0 sponses. where Nearly and Nlate are the number of samples in the Real RIRs were captured in various environments, including: early and late segments, respectively. • Faculty classrooms • Offices 2.2.5 Mel Spectrogram • Residential premises In addition to the features already discussed, a Mel spectrogram of the audio data is also computed for use in Convolutional Neu- • Concert halls ral Networks (CNNs). The power spectrogram is converted into • Theaters decibels (dB) and its dimensions are expanded to fit the input • Music studios requirements of the CNN. • Musical direction areas The first 2.5s of a real RIR are shown in Fig. 1. SpectrogramdB = 10 · log (Spectrogram 10 Power) (12) The expanded dimensions ensure compatibility with the CNN’s architectural requirements, providing an enhanced fea- ture set for more robust classification or regression tasks. It is worth noting how a Mel spectrogram differs from a reg- ular spectrogram in several key aspects: • Linear vs Non-linear Frequency Scaling: A regular spec- trogram employs linear frequency scaling, where all fre- quency bins are equally spaced. On the other hand, a Mel spectrogram utilizes the Mel scale, which is a non-linear scale that mimics the human ear’s response to different frequencies. Fig. 1. First 2.5s of a real RIR. • Uniform vs Variable Resolution: A regular spectrogram offers uniform frequency resolution across the spectrum. Synthesized Impulse Responses were generated using the ODEON software. This software models a variety of spaces, in- In contrast, a Mel spectrogram has higher resolution at cluding: lower frequencies and lower resolution at higher fre- quencies, better aligning with human auditory percep- • Parallelepiped rooms tion. • Churches • FFT-Based vs Filter Banks: A regular spectrogram is often • Concert halls derived directly from the Fast Fourier Transform (FFT) of • Rooms with irregular shapes the signal. In contrast, a Mel spectrogram is usually ob- tained by applying a set of Mel filters to the power spec- The first 2.5s of a synthetic RIR are shown in Fig. 2. trum of the signal. About ODEON Software: ODEON is a renowned tool in spa- tial acoustics, used for both simulations and measurements. It is • Psychoacoustic Properties: Due to its design, the Mel applied in analyzing sound fields in enclosed environments, such scale in a Mel spectrogram captures psychoacoustic prop- as theaters, concert halls, and music studios, as well as more erties that are more in line with human perception, mak- expansive spaces like stadiums. The software employs a combi- ing it more useful for tasks like speech and audio recog- nation of ray tracing models and image-source methods for its nition. simulations. Pap et al.: Real or Synthetic? A Machine Learning Approach to Classifying Room Impulse Responses 30 Fig. 2. First 2.5s of a synthetic RIR. With ODEON, users can design rooms of any shape and size, and subsequently inspect the key sound field characteristics at desired points within the room. Its versatility makes it a go-to choice for predicting acoustic profiles in large areas such as concert halls, airport terminals, and metro stations. Furthermore, ODEON introduces a specialized ray tracing algorithm tailored for noise prediction. Generating an acoustic profile within a room using ODEON Fig. 3. Confusion Matrix of the Dual-branch Neural Net- requires a geometric model of that room. Users can either work Classifier. create this directly within ODEON or import it from 3D draw- ing applications like AutoCAD. Once the model is created, ODEON offers tools for visual model validation to ensure its the model’s high accuracy. completeness. Any discrepancies detected can be quickly ad- dressed. Every model is comprised of surfaces, each asso- Metric Value ciated with a specific material. These materials, character- Accuracy 0.9939 ized by frequency-dependent absorption coefficients, are well- Balanced Accuracy 0.9944 represented in ODEON’s extensive database. This database en- Recall 0.9944 compasses almost all conceivable materials found in real-world Selectivity 0.0056 scenarios. Additionally, users have the flexibility to expand the F1 score 0.9939 database by adding new materials and defining their unique ab- sorption coefficients. Table 1. Performance metrics of the dual-branch neural Our dataset, diverse in nature, encompasses a broad spec- network classifier. trum of both real and synthesized spaces. With 127 synthetic samples and 267 real samples, it serves as the foundation for our research, aiming for robust and generalized results. Table 1 represents the performance metrics of the dual- branch neural network classifier. The accuracy, which is the pro- 3. RESULTS portion of instances that the model correctly predicted out of all instances, stands at 0.9939, suggesting that our model was able This heatmap shown in Fig. 3 illustrates the classifier’s perfor- to correctly classify about 99.39% of the instances in the test mance on the test set. The X-axis represents the predicted la- data. Balanced accuracy, the average of recall obtained on each bels and the Y-axis represents the true labels. The values in the class, stands at 0.9944. This high score indicates the model’s re- matrix indicate the number of instances for each combination liability and effectiveness in identifying both real and synthetic of predicted and true labels. The high values on the diagonal in-room impulse responses. Recall, also known as sensitivity or dicate that the classifier correctly identified a large majority of true positive rate, is a measure of how well the model identified both real and synthetic room impulse responses, with 88 true real room impulse responses. With a score of 0.9944, the model positives and 75 true negatives. The off-diagonal values indicate correctly identified 99.44% of the real room impulse responses. misclassifications, with only one instance of a real response in- Selectivity, or the true negative rate, is at a minimal 0.0056, correctly classified as synthetic (false positive), demonstrating meaning the model misclassified only 0.56% of synthetic room Pap et al.: Real or Synthetic? A Machine Learning Approach to Classifying Room Impulse Responses 31 impulse responses. Finally, the F1 score, which provides a bal- challenge and refine our model’s classification capabili- ance between precision and recall, is at a near-perfect 0.9939, ties. indicating that the model achieved an exceptional balance be- 2. Model Refinement: The model’s performance in spe- tween these two crucial metrics. cific applications, such as audio forensics or virtual real- ity, could be investigated and optimized. This would in- 4. CONCLUSION volve tailoring the model to the unique requirements and constraints of these applications, which could lead to im- The dual-branch neural network classifier we presented in this provements in model performance and practical utility. paper has shown exceptional performance in distinguishing be- 3. tween real and synthetic room impulse responses, as demon- Feature Set Expansion: Exploring ways to incorporate other types of data and feature sets into our model can strated by our results. This performance not only adds credibil- offer new perspectives and enhance the model’s perfor- ity to the technique but also opens new avenues for acoustic and mance. Future research could consider the inclusion of audio research, where such distinction plays a pivotal role. additional or alternative acoustic features, or even con- Our model achieved an impressive accuracy rate of 99.39%, sider other forms of input such as spatial or temporal demonstrating its robustness in classifying room impulse re- data. sponses. This high accuracy indicates that the dual-branch ar- chitecture, utilizing both a Convolutional Neural Network (CNN) 4. Model Interpretability: As our model becomes more and a Fully Connected Network (FCN), allows for comprehensive complex, understanding why it makes certain predic- analysis of data. The combination of time-frequency character- tions can be as important as the predictions themselves. istics derived from spectrograms processed by the CNN branch, Future work could include developing methods to bet- along with a set of acoustical features processed by the FCN ter interpret the model’s predictions and understand its branch, was clearly effective. decision-making process. Moreover, the model achieved a balanced accuracy score In taking these future steps, we will continue to refine and of 99.44%, demonstrating its ability to avoid bias and maintain expand the applications of our dual-branch neural network clas- a high level of performance across different classes. This is par- sifier, contributing to the growth and development of the field ticularly important in applications where the distribution of real of audio and acoustic research. and synthetic room impulse responses can vary widely. Despite the high performance achieved by our model, it is essential to consider these results within the scope of the cur- 6. REFERENCES rent study. The dual-branch classifier’s efficacy and efficiency [1] Pätynen, J. and Lokki, T. Evaluation of Concert Hall Auraliza-have been proven with our specific dataset and chosen acousti- tion with Virtual Symphony Orchestra, The Journal of the cal features. The results may differ if we use other types of data Acoustical Society of America, 101(6), pp.3517-3524, 2011. or feature sets. Future studies may further validate and gener- alize the utility of our model by testing it on other datasets and [2] Deppisch, T. and Gar´ı, S. V. A. and Calamia, P. and Ahrens, using other acoustic features. J. Direct and Residual Subspace Decomposition of Spatial The implications of our model extend to several domains. Room Impulse Responses, IEEE/ACM Transactions on Audio, In audio forensics, verifying the authenticity of room impulse Speech, and Language Processing, 31, pp. 927-942, 2023. responses can aid in the detection and prosecution of audio [3] Mi, Huan and Kearney, G. and Daffern, H. Impact Thresholds forgery. In acoustic environment modeling and virtual reality, of Parameters of Binaural Room Impulse Responses (BRIRs) ensuring the accurate representation of room acoustics is crucial on Perceptual Reverberation, Applied Acoustics, 134, pp. 1- for user experience and system performance. We believe our 7, 2022. model offers a novel, accurate tool to maintain the integrity and [4] Ratnarajah, Anton et al. Fast-Rir: Fast Neural Diffuse Room authenticity of room impulse responses in these applications. Impulse Response Generator, IEEE International Confer- ence on Acoustics, Speech and Signal Processing (ICASSP), pp. 571-575, 2022. 5. FUTURE WORK [5] Pezzoli, Mirco et al. Deep Prior Approach for Room Impulse While the results of our study are promising, there are several Response Reconstruction, Sensors, 22(7), p.2710, 2022. areas for further research and improvement: [6] Dilungana, Stéphane and Deleforge, Antoine and Foy, C. and 1. Dataset Validation: Further validation of our model on Faisan, S. Geometry-Informed Estimation of Surface Ab- a variety of datasets will contribute to generalizing the sorption Profiles from Room Impulse Responses, 30th Eu- model’s utility. Different types of room impulse re- ropean Signal Processing Conference (EUSIPCO), pp. 867- sponses and acoustic environments could be used to 871, 2022. Pap et al.: Real or Synthetic? A Machine Learning Approach to Classifying Room Impulse Responses 32 [7] Yu, Wangyang and Kleijn, W. Room Acoustical Parameter Es- method for ADHD based on brain network analysis of timation From Room Impulse Responses Using Deep Neu- resting-state fMRI images and transfer learning neural net- ral Networks, IEEE/ACM Transactions on Audio, Speech, and work, Frontiers in Human Neuroscience, 2022. Language Processing, 29, pp. 436-447, 2020. [10] Kayan, Ceyhun Efe et al. Intensity and phase stacked anal- [8] Foy, C. et al. Mean absorption estimation from room im- ysis of a Φ-OTDR system using deep transfer learning and pulse responses using virtually supervised learning, The recurrent neural networks, Applied Optics, 2022. Journal of the Acoustical Society of America, 150(2), pp.1286- [11] Simonyan, Karen and Zisserman, Andrew. Very deep convo- 1299, 2021. lutional networks for large-scale image recognition, arXiv [9] Meng, Xiaojing et al. Diagnostic model optimization preprint arXiv:1409.1556, 2014. Pap et al.: Real or Synthetic? A Machine Learning Approach to Classifying Room Impulse Responses 33 LCA STUDY OF DIFFERENT RECYCLED SOUND ABSORBERS FROM MELAMINE FOAM WASTE Urban Kavka1, Alberto Quintana-Gallardo2, Martina Marija Vrhovnik1, Rok Prislan1 1 InnoRenew CoE, Livade 6a, Izola, Slovenia (urban.kavka@innorenew.eu) 2 Center for Physics Technologies, Universitat Polit ècnica de Val ència, Val ència, Spain Abstract: facturing elements from renewable materials or recycling and The main focus of this study was to investigate the use of reusing waste materials to extend the life of already embodied melamine foam waste in the production of sound absorption energy in the materials. The latter is conventionally overlooked panels, with emphasis on performing a life cycle assessment because it is very time consuming, but could be a good option analysis (LCA) for two different types of sound absorbers. LCA if higher quality waste materials are recycled. One such exam-Impacts were studied for panels made from rigid monolithic ple is explored in this paper, where melamine foam waste from melamine foam and from randomly cut fragments. Most impor- the production of anechoic chamber at InnoRenew CoE was re- tantly, the different shape requirements can greatly affect the cycled into sound-absorbing panels meant for lecture halls and recycling potential of the material as well as the acoustic and offices. The melamine foam waste is of higher quality due to its environmental performance. Environmental impacts were cal- sound absorbing properties, and it makes sense to reuse it in this culated for different units. It has been shown that per surface case. This was done in two ways; producing sound panels from area of sound absorbing panel, which is 0.675 m2, the difference monolithic foam blocks and using melamine foam fragments ob-between the two types can be misleading. Therefore, evaluating tained by grinding smaller leftovers. The aim of this study is to the environmental impact per sound absorption area in each oc- upgrade comparison of an acoustic performance of two differ- tave band provides greater potential for a holistic comparison of ent types of sound insulation panels by assessing and comparing performance. Along with the acoustic performance studies, the their environmental performance using a life cycle assessment overall performance and impact of each type was determined, (LCA). with the monolithic type having better overall performance up A comparison of sound-absorbing performance was made to 250 Hz and the fragment absorber above. The study provides in a previous study [2], but a calculation error was later discov- important insight into the evaluation process by offering a novel ered and the corrected values are presented here. The calcula-comparison of overall performance, PAE . tion error occurred in the calculation of the ratio of melamine Keywords: LCA Study, reusing waste foam, sound absorbers, foam to voids in the case of the sound absorber type with loose P fragments. This ratio was calculated as 2:1, while in reality it is AE 1:2, which means that there are about 33 % of melamine foam and 67 % of air cavities. This error does not necessarily affect 1. INTRODUCTION the performance of the panels, but the analysis of the results The building sector generally needs more sustainable solutions, and the discussion is renewed in this paper. as the use of buildings accounts for between 30 and 40 per- cent of total global energy consumption and consequently be- 2. METHODOLOGY tween 40 and 50 percent of all greenhouse gases [1]. However, not all building components and design areas are pressured into 2.1 Subject of study a change equall; among the less affected is building acoustics. Even though its material contribution to the overall building is In this study, two types of sound-absorbing panels with differ- relatively modest, the drive to reduce the environmental foot- ence in shape of absorption material were analysed; the figure print should be as strong in this particular sub-sector as in oth- 1 shows the monolithic type with blocks of melamine foam, and ers. the figure 2 shows the absorber with loose melamine foam frag- Two equally important steps can be taken to obtain acous- ments. tic elements with lower environmental impacts: either manu- For LCA purposes, a list of materials used in the production 1st Author Surname et al.: Paper title 34 Material Monolithic type Fragment type Wooden frame 4,05 dm3 4,05 dm3 (fiberboard) Fabric (front, 180 g 360 g 100 % PET) (Single-layered) (Double-layered) Fabric (back, 150 g 150 g 100 % PP) Melamine foam 255 g 90 g Screws 8 pieces 8 pieces Staples 80 pieces 80 pieces Table 1. List of materials used for the production of one sound absorbing panel for each type. Fig. 1. Monolithic type of a sound absorber of the product. This can be considered as ”cradle to gate” LCA. One reason for limiting the scope of this study is that cradle-to- gate LCA is considered appropriate for this study, as the possible end-of-life scenarios of both panels can be considered similar, as both use the same raw materials. 2.2.1 Functional unit The functional unit in a study of LCA refers to the element used as a comparative reference. In this case, the functional unit is the production of a 0.675 m2 acoustic panel for both types. 2.2.2 Allocation principle The allocation principle used in this study is end-of-life (EoL) allocation according to EN 15804:2020 [4]. The methodology was adapted from Baldassarri et al. [5] and Lavagna et al. [6] imple- mented. Fig. 2. Fill-in fragment’s type of a sound absorber 2.2.3 Life Cycle Inventory The software used to create the LCI was Simapro v9.5. of both types of sound-absorbing panels was prepared and is Simapro incorporates Ecoinvent V3.9.1, the most comprehensive presented in the table 2.1. The main difference between the two database for LCA calculations [7]. The data extracted from Ecoin- types of panels is the amount of melamine foam used and the vent represent the average European production of the studied amount of front textile. processes. 2.2 LCA 3. RESULTS A comparative LCA has been performed to assess the differences in environmental impacts of a 0.675 m2 acoustic panel made The Environmental Footprint characterization results, given in with monolithic blocks of waste melamine foam and a similar the table 3, show that the fill-in fragments panel reduces the en- panel made using fragments of waste melamine foam with frag- vironmental impact of each studied category. For some impact ment sizes up to ϕ3.0cm. categories, the impacts are only slightly lower, such as climate The LCAs were performed according to the guidelines de- change potential, which is reduced by 20 %, and acidification, scribed in ISO 14040:2006 [3] and EN 15804:2020 [4]. The mod- which is reduced by about 27 %. In the case of water consump- ules considered are A1 to A3, which takes into account all pro- tion, the water consumption required to manufacture a panel is cesses from the extraction of raw materials to the manufacture reduced by 70 %. 1st Author Surname et al.: Paper title 35 Fig. 3. Environmental footprint results, firstly normalized (upper graph) and then weighted (bottom graph) 1st Author Surname et al.: Paper title 36 Damage Monolithic Fragment Unit the sound absorption coefficients are the key to decision making category type type and are shown in the figure 4. Acidification mol H+ eq 0.0180 0.0127 Climate kg CO2 eq 3.20 2.57 change Ecotoxicity, CTUe 0.0079 0.0083 freshwater Particulate disease inc. 3.1905 2.5585 matter Eutrophication, kg N eq 0.0028 0.0028 marine Eutrophication, kg P eq 7.9059 6.7599 freshwater Eutrophication, mol N eq 4.7943 4.0539 terrestrial Human toxicity, CTUh 11.3550 9.4744 cancer Human toxicity, CTUh 0.3900 0.4077 non-cancer Ionising kBq U-235 eq 0.9552 0.9316 radiation Land use Pt 2.33E-07 1.64E-07 Ozone depletion kg CFC11 eq 3.09E-03 2.54E-03 Photochemical Fig. 4. Sound absorption coefficients for monolithic kg NMVOC eq 1.50E-04 1.50E-04 ozone formation melamine foam panels and panels with fill-in melamine Resource use, MJ 4.04E-02 2.71E-02 foam fragments in 1/3-octave bands fossils Resource use, kg sb eq 1.03E-08 2.08E-09 minerals, and metals As can be seen in the figure, the fragment absorber curve Water use m3 depriv. 1.32E-09 1.10E-09 shifts to the higher frequencies for about two 1/3-octave bands. Considering the final use of the panels in lecture halls and of- Table 2. Environmental footprint characterization results fices, the focus of the study was narrowed down to octave bands for the monolithic and fragment type of sound absorbing between 125 Hz and 4000 Hz, covering the octave bands of the panel. human voice and the highest sensitivity of human hearing. The objective of the study was to determine a single value representing both sound absorption and environmental perfor- mance, although frequency dependence could not be avoided. In the upper graph of Figure 3, the normalized results reveal Since an increasing value of α improves the performance of the a high impact on Resource Use in both panels. There is a sig- panel, and a decrease in the single score result of LCA also im- nificant difference between the human toxicity category in the proves the performance, the following equation for calculating two panels. The main reason for this result is the toxins associ- a single value performance, denoted by PAE, was proposed to ated with the urea production, which is necessary to manufac- merge the two characteristics: ture melamine foam. According to the weighting (see lower di- α agram in Figure 3), the categories Climate Change and Resource PAE = (1) LCAsingle score Use are those with the higher relative importance. Considering , that the results are normalized and weighted, the environmen- where LCAsinglescore is the single score result calcu- tal footprint offers the possibility of obtaining a single impact lated as the sum of the weighted and normalized significant pa- score result by adding up each category. The single score result rameters of the LCA study, and α is the sound absorption coeffi- for the monolithic panel is 326.22 µPt, while that for the fill-in cient in each octave band of the respective product. The calcu-fragments is 244.70 µPt. lated values of PAE for the monolithic and fragmented types are shown in the table 4, where a higher value represents a panel with better overall performance. 4. DISCUSSION As can be seen in the table 4, the monolithic panel type gen- Selection of the most suitable sound absorbers focuses on the erally has better overall performance up to the 250 Hz octave acoustic performance of the panels, which is later influenced by band and the fragment type of a panel has better overall perfor- the LCA study or the comparison of other aspects. Therefore, mance above 250 Hz. 1st Author Surname et al.: Paper title 37 P duction of other products, hence their influence on the overall Frequency [Hz] AE Monolithic Fragment environmental footprint could be disregarded. Such a principle 63 0.00 4.09 × 10−5 could be applied in our study, but it would significantly change 125 1.84 × 10−4 8.17 × 10−5 the results of the LCA study. Consequently, the influence of pro- 250 1.23 × 10−3 8.17 × 10−4 cesses would be higher, so it would be important to consider the exact differences in the processes required and their resources 500 2.54 × 10−3 2.82 × 10−3 for the production of each absorber type. 1000 2.64 × 10−3 3.68 × 10−3 For a holistic assessment of the overall performance of ab- 2000 2.48 × 10−3 3.47 × 10−3 sorbers, the production processes and their environmental foot- 4000 2.48 × 10−3 3.47 × 10−3 print should be included, along with an extension of the LCA study to cradle-to-cradle scope. Most importantly, different Table 3. PAE values for both types of panels in each de- shape-related requirements can largely influence the recycling pendency of the frequency. potential of the material. Therefore, it would be important to ask how the reusability and recyclability, especially of the fill-in fragment panel type, can be assessed or even quantified. The proposed single-value characterization of overall acous- There are also two ways to ease this comparison. First, the tic and environmental performance, PAE, is a novel step to- thickness of the fill-in fragment type could be increased so that wards a holistic evaluation of acoustic elements to not only ob- the curve of the absorption coefficient in the figure4 would shift tain desired acoustic result, but also to contribute to the reducto lower frequencies to the point where the curves of both ele- tion of environmental impact in the building sector. Neverthe- ments would coincide. In this way, it could be assumed that the less, the proposed single-valued performance quantifier needs acoustic performance of both types is the same. Based on the further development and research as it is still frequency depen- modified dimensions of the acoustic panels, a new LCA study dent. Another obstacle to be overcome is that the acoustic per- would then be carried out, the results of which would already formance is normalized and quantified between 0 and 1. On the be suitable as individual values for comparing the overall per- other hand, the single score result of the LCA study depends on formance. the selection of significant parameters and has no upper limit. Second, a fill-in fragment type could be modified so that the Hence dividing α by the LCA single-score result yields small num- consumption of melamine foam would be the same for both ele- bers, and the difference between the scores of the different sub- ments. Assuming that the consumption of other materials does jects in the comparison does not necessarily tell us about the not change significantly, or that a change in their quantity does actual difference in overall performance. It could also be mod- not significantly affect the LCA study, an assumption of equal ified to consider only a single category of environmental foot- environmental footprint results could be made. Thus, acoustic print category, such as water use.Another way to overcome this performance would be the main subject of performance com- obstacle, instead of using PAE, is to change the dimensions of parison. However here it is important to check for the effect the panels to match either the acoustic performance or the LCA of each material used and determine if melamine foam has the results, and then compare the other. Nevertheless, the novel greatest impact on the environmental footprint. integration of LCA and acoustic performance is recognised as a potential area for future research. 5. CONCLUSION The fill-in fragment panel reduces the carbon footprint by 20 %, 6. REFERENCES which is similar to reduction in the absorption coefficient in the frequency range 125 and 500 Hz, where the difference between [1] T. Ramesh, R. Prakash, and K. Shukla, “Life cycle energy anal- the two absorbers is largest. According to the individual results ysis of buildings: An overview,” Energy and Buildings, vol. 42, obtained by the Environmental Footprint methodology, the to-pp. 1592–1600, 10 2010. tal environmental impact is reduced by 25 % in the case of the [2] U. Kavka, M. M. Vrhovnik, and R. Prislan, “The potential of fragmented absorber type compared to the monolithic type. melamine foam waste: Performance comparison of mono- Finding ways to reuse materials is crucial for a more sus- lithic foam block and loose foam fragments used as a filler tainable construction industry. In terms of environmental per- for sound absorbing panels,” 2023. formance, these materials have two main advantages: first, the [3] Standard, “Iso 14040:2006 environmental management — amount of waste is reduced due to a higher recycling rate, and life cycle assessment — principles and framework,” 2006. second, the use of new raw materials and the impacts associated with the production process of the new product are avoided. In [4] C. E. N. EN, “15804: 2012+ a2: 2019—sustainability the case of the sound absorbers studied, the melamine foam of construction works—environmental product declara- and the PP fabric were used as waste materials from the pro- tions—core rules for the product category of construction 1st Author Surname et al.: Paper title 38 products,” European Committee for Standardization (CEN): impact of housing in europe: Definition of archetypes and Brussels, Belgium, 2019. lca of the residential building stock,” Building and Environ- ment, vol. 145, 2018. [5] C. Baldassarri, K. Allacker, F. Reale, V. Castellani, and S. Sala, Consumer Footprint. Basket of Products indicator on House- [7] G. Wernet, C. Bauer, B. Steubing, J. Reinhard, E. Moreno- hold goods. 2017. Ruiz, and B. Weidema, “The ecoinvent database version 3 (part i): overview and methodology,” International Journal [6] M. Lavagna, C. Baldassarri, A. Campioli, S. Giorgi, A. D. Valle, of Life Cycle Assessment, vol. 21, 2016. V. Castellani, and S. Sala, “Benchmarks for environmental 1st Author Surname et al.: Paper title 39 MULTICHANNEL REVERBERATION TIME MEASUREMENTS OF MIURA-ORI ORIGAMI IN ALPHA CHAMBER Jurij Prezelj, Andrej Hvastja, Ema Letnar, Miha Brojan Faculty of mechanical engineering, University of Ljubljana Abstract: In this research, we introduce and validate a novel LabVIEW program tailored to evaluate sound absorption characteristics of materials in custom made alpha chamber, specifically for Miura-Ori origami structures crafted from thick paper. Initial validation involved simulating sound dissipation signals, using harmonic signal with an exponential function representing sound energy decay. Software validation was performed by matching synthesized theoretical signal with the program's outputs. Practical validation was demonstrated using a stone wool sample, with known sound absorption from interlaboratory comparison. During validation a hypothesis that progressively increasing the number of reverberation time evaluations in the alpha chamber, averaged result drifts closer to its inherent reverberation time. The purpose of new LabView program for multichannel measurements of reverberation time and sound absorption is to observe a relationship between material/structure acoustic attributes and experimental setups, and to confirm the hypothesis that a significant correlation exists between the sound absorption frequency and the spatial frequency of the Origami surface. This hypothesis was placed after a reactive behavior in the sound absorption mechanism of Miura-Ori Origami structure was identified while observing a correlation between the sound absorption spectrum and the spatial Fourier transformation spectrum of the origami structure surface. Keywords: Reverberation Time, Sound Absorption Coefficient, Spatial frequency, Two-dimensional Fourier transformation, Alpha Chamber, Diffuse Sound Field, Resonant Absorption, Multichannel Software 1. INTRODUCTION Origami, the Japanese art of paper folding, is renowned for its distinctive mechanical attributes and Noise control is vital in numerous engineering adaptability. Beyond art, its principles can inspire the applications, from designing quiet vehicle interiors to creation of versatile, reconfigurable structures apt for developing noise-dampening building materials, various engineering feats. In acoustics, such structures advanced communication technologies, and medical hint at groundbreaking sound-absorbing materials and devices. A successful noise control strategy hinges on a devices. These devices can dynamically alter their form profound understanding of the sound absorption and and functionality, enabling real-time modulation of resonance characteristics of the involved materials and acoustic properties. Our research therefore bifurcates structures. into two interconnected segments: Traditionally, sound management has relied on 1. Development and Validation of Measurement materials with static geometric properties. Once Procedure: We introduce an innovative multichannel manufactured and installed, their acoustic properties procedure to determine the reverberation time (T) within remain fixed. This rigidity has prompted researchers to a compact reverberation chamber. Our methodology seek adaptive, flexible structures capable of dynamic circumvents the hurdles tied to capturing precise T sound absorption and directionality control, entering the measurements in imperfectly diffusive acoustic world of origami-inspired structures. environments. We achieve this by amplifying the 40 AAAA – 2023 – IZOLA - Conference Proceedings measurement volume and enhancing their spatial o Diamant et al. innovated an acoustic tank with Miura- distribution. Ori patterned absorption plates [8], while Zou et al. 2. Exploration of Origami Mechanics in Acoustics: investigated arrays rooted in origami tessellations Delving deep into the mechanics of origami structures, we [9,10]. probe their sound absorption capabilities. Our objective is o Modular origami as adaptable silencers were studied twofold: to recognize potential acoustic resonances and by Fang et al. and Zhu et al., with a focus on tunable to assess the adaptability of origami for acoustic sound attenuation [2, 11]. applications. We harbor a keen interest in understanding o Jiang et al. unveiled an origami-based sound absorber the relationship between the two-dimensional Fourier displaying adjustable resonance frequencies, with a transformation of the origami surface and its sound noticeable average sound absorption coefficient [3]. absorption coefficient. o Cambonie et al. employed origami spirals in compact quarter-wavelength resonators, simulating the 2. ORIGAMI IN ACOUSTICS absorption properties of longer designs [12]. Origami, inspired by the ancient art of paper folding, Origami and Wave Energy: The periodic nature of has stimulated innovation across numerous engineering origami structures, with inbuilt bandgap features, can domains, from robotics and biomechanics to architecture effectively guide wave energy [13]. The potential to adjust and aerospace [1]. Fonseca et al. emphasized this trend, waveguide frequency, especially during transitions noting that its full potential, especially in acoustics, is still between lattice types, underscores the adaptability of untapped [1]. origami in acoustic applications [13]. Benefits of Origami Designs: Origami structures bring Considering the promising directions identified in about transformative capabilities like structural existing literature, our research sought to bridge a gap, by reconfigurability, acoustic tunability, and design flexibility. placing a hypothesis: Such features address current shortcomings in sound Hypothesis 1: A significant correlation exists attenuation devices, especially the limited adaptability of between the sound absorption frequency and the spatial silencers to multiple operational scenarios [2]. frequency of the Origami surface. Broadband Acoustic Absorption: The pursuit for We conceptualized a multichannel measurement effective absorption across a wide frequency range is system (Alpha chamber with the fitted LabVIEW program) paramount. Yet, many contemporary methods face for speeding up the assessments of sound absorption. challenges, such as bulky designs or narrow absorption With most of the research emphasizing impedance tube bandwidths. Origami structures, however, shine by measurements, a void remains concerning origami widening the sound absorption spectrum and bolstering structures' sound absorption evaluations in real-world low-frequency sound absorption — a promising avenue settings, particularly in diffuse sound fields. This for controlling low-frequency noise [2, 3]. experimental pursuit is made to understand the origami Innovative Origami Applications: structures' sound absorption abilities. o J. Xiaomeng et al. developed origami windows whose opening area can be adjusted by a single-degree-of- 3. MEASUREMENTS OF SOUND ABSORPTION freedom folding mechanism, achieving significant transmission loss while maintaining ventilation [44]. Sound absorption materials are critical in diverse noise o Harne et al. highlighted the utility of a Miura-Ori-based control scenarios, spanning room acoustics to automotive acoustic array for focused acoustic energy, while interiors. Standard practice often measures their Pratapa et al. engineered origami-inspired attributes using impedance tubes and small samples, metamaterials with notable acoustic switching working under the presumption of a one-dimensional potential [5, 6]. sound field. Yet, these conditions seldom mimic real- o Yu et al. and Xiang et al. explored reconfigurable 3D world scenarios, where materials of varying sizes and sound barriers using the Miura-Ori pattern, shapes are used in multidimensional sound fields. emphasizing their advantages over traditional barriers In response to this discrepancy, the reverberation in terms of adaptability, portability, and cost [7]. chamber, in line with the international standard ISO 354, 1st Author Surname et al.: Paper title 41 AAAA – 2023 – IZOLA - Conference Proceedings 2003 [14], has emerged as an alternative. However, distribution of absorption throughout the room enhanced crafting these expansive chambers for absorption the level of diffusion. They observed that measurements coefficient evaluations is both costly and labor-intensive. taken in a slightly larger room with non-parallel walls A more recent solution has been the introduction of alpha showed that the difference between untreated and chambers, compactly sized at roughly 6.4 m³. treated walls was not so pronounced. This indicates that Still, alpha chambers are not without limitations. Their non-parallel walls improved the diffusion level, regardless performance tends to falter, particularly at lower of wall treatment. frequency ranges, where a distinct shift from the standing Despite large amount of literature, there is no study sound field to the diffuse sound field becomes apparent. available in which necessary number of measurements In such scenarios, factors like the chamber's geometry, would be proposed as function of the Relative Standard positioning of samples, location of the sound source, and Deviation (RSD), and to dive into the possibility to microphone placement can greatly skew absorption overcame problems of non ideal diffuse field by coefficient readings. significantly increasing the number of measurements. A diffuse sound field has uniformly distributed sound Driven by this insight, our research postulates the second energy where propagation directions are random at any hypothesis: point. Perfect diffusion is theoretical and used for Hypothesis 2: By progressively increasing the reverberation time calculations. Randall et al. [15], number of reverberation time evaluations in the alpha highlighted that poor absorption distribution reduces chamber, we inch closer to the chamber's true, inherent diffusion in specific frequencies. Adjusting room geometry reverberation time. can improve diffusivity but might negatively impact other Based on the outlined problem, we developed frequencies. Better surface absorption distribution or software for sound absorption measurements in small introducing artificial diffusers can enhance room reverberation chambers. This software offers multi- diffusivity. channel reverberation time measurements. Such multi- In the work of Cops et al. [16], they raised the channel measurement capability allows for simultaneous question, "Does the room's diffusivity improve with the signal capture from microphones at various positions, increase in the number of diffusers or by decreasing the considerably shortening the measurement procedure axial and tangential modal locations at the reverberation duration. Furthermore, the streamlined nature of the corners?" Figure 1 shows a test with one wall covered in process enables the execution of a larger number of absorbent material (condition A8), leading to uneven measurements. As the number of measurement points absorption. In condition A9, 12 diffusers were added, and increases, the measurements become less influenced by in A10, they were replaced with absorbent material. The the chamber's inherent characteristics, such as the overall room absorption in the frequency range remained challenge presented by an imperfect diffuse sound field, largely consistent. Diffusers significantly influence the which is a significant concern in the utilization of such room's absorption. systems. 3.1 Theoretical background The evaluation of acoustic properties within enclosed environments necessitates the understanding and accurate calculation of reverberation time (T). Reverberation time is a pivotal parameter, defined as the duration for the sound pressure level in a space to decrease by 60 dB after the cessation of a sound source. Figure 1: Reverberation time of the space under Its precise calculation offers insights into the acoustic conditions A8, A9, and A10, [15]. quality of spaces, from intimate living quarters to grand concert halls. For the calculation of reverberation times, When examining a room where absorption was we can use several methods presented in the work of unevenly distributed, placing diffusers and improving the Kinsler et al. [20]. Historically, researchers have employed 1st Author Surname et al.: Paper title 42 AAAA – 2023 – IZOLA - Conference Proceedings various methods to compute reverberation times. One of approached as follows. First, we calculate the equivalent the influential works in this field is that of Kinsler et al. absorption surface of the empty reverberation chamber [20], which presents a more methodologies. Among these, A 1 using the following equation: Sabine's and Eyring's methods are frequently employed due to their efficacy and historical significance. As 55,3 𝑉 𝐴 (1) 1 = − 4𝑉𝑚 𝑐𝑇 1 presented in the work of Beranek [17], Sabine's equation 1 is derived under the assumption that sound waves reflect where V is the volume of the empty reverberation sequentially from wall to wall. Another commonly used chamber in cubic meters, c is the speed of sound, T 1 is the method, the Eyring method, assumes that the initial reverberation time of the empty reverberation chamber waves reflect off all the walls of the room at once, with in seconds, and m 1 represents the coefficient for recording subsequent reflections reducing the sound energy by the the climatic effects on sound absorption in the air. This average absorption coefficient of the room. However, in changes with changes in temperature, humidity, and air our study the standardized EN SIO 354-2003 method was pressure in the reverberation chamber. The coefficient used. m1m1 is calculated using the following equation: In the EN ISO 354-2003 standard, two methods for performing T reverberation time measurements are 𝛼 𝑚 (2) presented: the interrupted sound source method and the 1 = 10𝑙𝑜𝑔𝑒 integrated impulse method. When using the signal interruption method, the standard provides the following The sound damping coefficient α is due to atmospheric recommendations: influences. As presented in the ISO 9613-1:1993 standard • Microphones used for reverberation time [14], this is calculated as follows: measurements should be omnidirectional. • Multiple measurements should be made with 𝑝 −1 𝑇 1 2 ⁄ 𝑇 −5 2 ⁄ (3) 𝛼 = 8,8686𝑓2 ([1,84 ∙ 10−11 ( 𝑎) ( ) ] + ( ) microphones spaced at least 1.5 m apart. They should 𝑝𝑟 𝑇0 𝑇0 −1 be at least 2 m away from the sound source and at least −2239,1 𝑓2 ∙ {0,01275 [𝑒 𝑇 ] [𝑓𝑟𝑂 + ( )] 𝑓 1 m away from the walls. 𝑟𝑂 −1 −3352,0 𝑓2 • The number of independent sound energy dissipation + 0,1068 [𝑒 𝑇 ] [𝑓𝑟𝑁 + ( )] }) 𝑓𝑟𝑁 measurements should be at least 12. At least three microphone setups and two different sound source The coefficients frO and frN represent the relaxation locations should be used for measurements. frequencies of oxygen and nitrogen. Their values are given • The noise playback time should be sufficient to fill the by the following equations: space with sound and achieve a balanced sound pressure level across the entire frequency range. Only 𝑝 0,02 + ℎ 𝑓 𝑎 (24 + 4,04 ∙ 104ℎ ) (4) then can the source be turned off. 𝑟𝑂 = 𝑝𝑟 0,391 + ℎ • The sound pressure level before turning off the source −1 3 ⁄ 𝑝 −1 2 ⁄ 𝑇 𝑎 𝑇 −4,170[( ) 𝑇 −1] (5) should be high enough that the measurement ends at 𝑓 0 𝑟𝑁 = ( ) (9 + 280ℎ ∙ 𝑒 ) 𝑝𝑟 𝑇0 least 10 dB above the ambient sound pressure level. To reduce measurement uncertainty, the standard Similarly, we calculate the values when using a sample. recommends averaging at least three measurements on The equivalent absorption surface of the reverberation each microphone. In calculations, it is important to chamber with the inserted sample A2 is calculated using consider that the starting point of the considered range is the equation (1) where the values V, c and m have the 5 dB below the initial sound level. The range should be same values, but T1 is replaced by T2 representing the within 20 dB. The lower limit of the range should be 10 dB reverberation time of the reverberation chamber with the above the background noise level. inserted sample. The equivalent surface of the absorption As mentioned at the beginning of the chapter, to material AT is then calculated using the equation: determine the absorption coefficient, the equivalent surface of the sample must first be calculated. The 1 1 𝐴𝑇 = 𝐴2 − 𝐴1 = 55,3𝑉 ( − ) − 4𝑉(𝑚2 − 𝑚1) (6) calculation of the equivalent surface of the sample is 𝑐2𝑇2 𝑐1𝑇1 1st Author Surname et al.: Paper title 43 AAAA – 2023 – IZOLA - Conference Proceedings With the calculated equivalent surface of the absorption material, the sound absorption coefficient α s is calculated using the equation (9), where S represents the floor or wall surface covered by the sample. 𝐴 𝛼 𝑇 (7) 𝑠 = 𝑆 3.2 Experimental setup Figure 3: Reverberant modal sites as influenced by audio signal frequency. The acoustic measurements for our study were conducted in a reverberation chamber, with a sound field volume of Considering the influence of various standing wave V=6.44 m3. As described by Vrtovec [18], the chamber has patterns on results, executing a higher number of a dimensional ratio of 1.4:1.2:1. The primary support measurements is imperative. This approach ensures structure, depicted in Figures 1 and 2, integrates wooden minimizing any discrepancies introduced by the natural and wood-steel composite beams. Both the interior and interactions of sound waves within the chamber, exterior surfaces of the chamber are lined with 2 layers of rendering results more accurate and representative. plasterboard, and recycled compressed polyurethane Our signal generation combines a cosine function foam offers the necessary sound insulation. (simulating harmonic sound wave propagation) and an exponential function (emulating sound energy decay over time). The reverberation time, T, represents the period required for a sound level to drop by 60 dB. For this study, a synthesized signal with a T set to 1s at 1000 Hz frequency was chosen. Figure 5 depicts the decline in sound pressure level during the validation process. Figure 2: Wood-steel composite beams detail [18] Figure 3: Load-bearing structure overview [18] Figure 4: Sound pressure level drop during validation Standing sound fields, observable across all frequencies as Table 1: Reverberation Times obtained during the shown in Figure 3, are marked by positional variances in Validation, performed on synthesized signal. their minima (represented as lines) and amplitude Central frequency 500 630 793,7 1000 1259,9 1587,4 [Hz] fluctuations (deduced from the color scale). The interplay Measured between the sample's positioning and the standing sound reverberation NaN NaN 1.0405 1.0112 1.0393 NaN time T [s] field's orientation is critical. The variation in standing wave patterns necessitates a higher measurement frequency, Table 1 shows the measured values of T for selected ensuring results are more accurate and indicative of the narrow range of central frequencies. A NaN (Not a actual sound behavior within the chamber. 1st Author Surname et al.: Paper title 44 AAAA – 2023 – IZOLA - Conference Proceedings Number) value is noted where the level calculation was to drift and finally stabilizes to constant value. Transition unsuccessful, due to either the absence of the signal in depends on frequency. Value of T for Frequencies above that frequency range or a too low signal-to-noise ratio. 300 Hz is stabilized much sooner, while below 300 Hz a noticeable drift is observed even after 35 measurements. 4. MEASUREMENTS RESULTS 4.1 Reference Measurement Selected reference sample for validation of the method is stone wool, a non-flammable fibrous material widely recognized for its thermal insulation capabilities, largely attributed to its sound isolation qualities. With dimensions of 0.6 m x 1 m x 0.06 m, the sample has a width-to-length ratio of 0.6. When we account for side Figure 5: Drifting of the averaged reverberation time surfaces, its total surface area is 0.984 m2. This results in a value to its inherent value with increasing number of ratio E=5.3 between the perimeter and surface area of the measurements, for the reference measurement of the sample. This sample was particularly appealing for our empty alpha chamber. study as it had undergone previous reverberation time measurements, has therefore known values of sound absorption, obtained from interlaboratory comparisons, using traditional methods, making it an ideal reference point for our research. The reference measurement was conducted in the same alpha chamber as the subsequent measurements. The reference method is considered the commercial method provided by the instrument Norsonic Nor-140. As shown in Figure 5, 39 reverberation time measurements were taken in the empty alpha chamber, while 19 were recorded with the stone wool sample present. The Figure 6: Drifting of the averaged reverberation time sample, comprising two stone wool pieces described value to its inherent value with increasing number of earlier, was positioned at the chamber's center floor. measurements, for the reference measurement of in the These pieces were stacked, resulting in dimensions of 0.6 presence of a sample. m x 1 m and a thickness of 0.12 m. The sample was angled at 45° relative to the entrance plane. Throughout the We closely examined the length of the signal measurements, only the microphone positions varied; the playback, impacting the data points captured for sound source and sample remained constant, with the determining the sound level decay curve. We conducted source located at the bottom-left corner of the alpha 80 reverberation time measurements in the alpha chamber. Graphical representations are provided, chamber: both when empty and with a sample shown in illustrating the stabilization of the reverberation time Figures 7 and 8, which was a 0.6 m x 1 m stone wool value T with increasing number of measurements, and layered configuration angled at 45° to the entrance plane. hence drifting the results to its true value, Figures 5 to 8. While we adjusted microphone and sound source These stabilization graphs were constructed in such a positions throughout the experiment, the sample way that the n-th measurement of the reverberation time remained stationary. Playback durations varied between represents the average of all previous measurements. At a shorter 4-second span and a longer 9-second span, with lower frequencies, there's visible "creep," potentially noise filtered between 80 Hz and 10,000 Hz. indicating a too-small number of measurements. With an increased number of measurements, the creep is changed 1st Author Surname et al.: Paper title 45 AAAA – 2023 – IZOLA - Conference Proceedings Crucially, the outcomes from the 80 measurements using A our methodology and software were consistent, displaying results within ±5% of sound absorption for frequencies above 300 Hz, in line with the reference method. Our system displayed superior precision in measuring reverberation time T below 300 Hz compared to the reference method. Furthermore, the use of our advanced software ensured that the time required for the 80 measurements (consisting of 10 repetitions with 8 microphones) was equivalent to that of 20 repetitions using the reference system with a single microphone. This highlights the enhanced efficiency and precision of our B approach. As the number of measurements increased, clear variations appeared, indicative of the stabilization of reverberation times. This pattern supports the primary hypothesis: increasing data points for reverberation time measurements progressively approximates the true reverberation time inherent to the alpha chamber. 4.2 Precision assessment - Probing with a plain Paper Figure 7: Stabilization of the reverberation time values for the shorter (A) and longer (B) playback time A plain paper, sized 0.60 m x 0.60 m, was placed measurement in the case of an empty alpha chamber. centrally on the alpha chamber's floor to evaluate measurement accuracy. In an unobstructed environment, 79 reverberation time readings were collected, compared A to 55 when the paper was present. Notable variations in the Relative Standard Deviation of Reverberation Time Measurement (RSD) were observed within the 100 Hz to 500 Hz frequency range. This suggests the possible presence of resonant absorption elements. The inconsistencies in the measurements might stem from the paper's uneven placement and its direct contact with the floor. The presence of an air gap could have made the paper act as a reflective surface. This emphasizes the importance of maintaining an air gap for resonant absorption materials, allowing for energy dissipation B through sample oscillations. However, Figure 9 demonstrates that the paper, when placed on the floor, primarily absorbs in a restricted higher frequency range, while deviations from zero in frequency range below 300 Hz can be attributed to the non diffuse field. We can say again that method and measuring system provide acceptable results in frequency range above 300 Hz and very good results in frequency range above 1000 Hz and Figure 8: Stabilization of the reverberation time values up to 5 kHz. for the shorter (A) and longer (B) playback timer measurement in the presence of a sample. 1st Author Surname et al.: Paper title 46 AAAA – 2023 – IZOLA - Conference Proceedings results below 300 Hz possess a deviation exceeding 0.1 s, thereby affecting the precision of the absorption calculations. Conversely, for frequencies above 500 Hz, the deviation is less than 0.05 s, which assures accurate results. Remarkably, a consistent frequency of the absorption peak is observed for the same type of folding. Figure 9: Coefficient of absorption of paper sheet 4.3 Measurement results of the origami structure Figure 10 displays samples of the Miura-Ori origami crafted from paper. Their corresponding absorption coefficient spectra are depicted in Figure 11. An analysis of these results reveals that different samples exhibit peak absorption coefficients at varying frequencies. Specifically, samples ORIG_1_1,2,3 achieve their highest absorption at 3125 Hz (as seen in Figure 11A), while samples ORIG_2_4,5,6, indicate their peaks absorption at 1600 Hz (Figure 11B). ORIG 1_3 ORIG 1_2 ORIG 1_1 ORIG 1+ ORIG 2+ ORIG 3+ D NOK OK ORIG 2_6 ORIG 2_5 ORIG 2_4 Figure 10: Samples of the Miura-Ori origami, made from paper are shown. Notably, the sound absorption measurements suggest that some samples may have limited absorption in the Figure 11: Sound absorption measurement for samples: frequency range below 300 Hz. This observed behavior A (ORIG 1_1, ORIG 1_2 and ORIG 1_3) could potentially stem from the higher RSD in this B (ORIGAMI 2_4, ORIGAMI 2_5 and ORIG 2_6) frequency range. The relative standard deviation of these C (ORIG 1+, ORIG 2+ and ORIG 3+) measurements is presented in Figure 11D. It's evident that and relative standard deviation RSD of reverberation time measurement, averaged for all measurements in D 1st Author Surname et al.: Paper title 47 AAAA – 2023 – IZOLA - Conference Proceedings To interpret the measured results, we considered j = 1, 2,…n. The 3D Cartesian coordinate system, along the impact of sound refraction. Consequently, the surface with its origin and orientation, is shown in Fig. 12(c). The of the origami structure was analyzed in detail, based on coordinate vector (x, y, z) of Vi,j can be given as: the analytical description of its surface, followed by a Fourier transformation analysis. Following hypothesis can ( 𝜂 𝑗 − 1) asin ( 𝐴) odd 𝑖 (16) 2 be proposed: There is significant correlation between 𝑥 = {( 𝜂 𝜂 𝑗 − 1) asin ( 𝐴) + 𝑏cos ( 𝑍) even 𝑖 Sound absorption frequency and spatial frequency of 2 2 Origami surface. η y = (i − 1)bsin ( 𝑍) (17) 2 5. CORRELATION BETWEEN THE SPATIAL FREQUENCY 0 for odd 𝑗 (18) AND SOUND ABSORPTION FREQUENCY 𝑧 = { 𝜂 acos ( 𝐴) for even 𝑗 2 The specimens studied in this research were flexural Twelve parameters a, b, ϕ, m, n, θA, θZ, ηA, ηZ, la, lb, lt have specimens and the Miura-Ori model was used to make been defined to specify the Miura-Ori pattern. The first them. The geometric relationships required to achieve the five parameters are the pattern constants because they Miura pattern were extracted from the reference [19]. remain fixed regardless of the folded shape. The last seven The Miura-Ori pattern is specified on an unfolded plane by parameters are the pattern variables because they parameters a, b, and ϕ. The angles between the planes are depend upon the shape of the fold. Among these, the six denoted as θA and θZ and the angles between the edges as independent geometric relations have been defined in ηA and ηZ and were used to determine the geometry of the Eqs. (10) through (15). This means that six independent folded shape, as shown in figure below. parameters are required to determine all the parameters of a unique Miura pattern. In a specific Miura pattern, these six parameters can be determined using five pattern constants and one pattern variable. Based on this equation the surface of the origami was constructed in MATLAB, as shown in Fig. 13. Figure 12: Miura-Ori geometry Table 4 Geometric characteristics of Miura-Ori specimens. In this model, the following six relationships are established between the parameters: (1 + cos η Z) (1 - cos η A) = 4cos2 ϕ (10) cos η A = sin2 ϕ cosθZ - cos2 ϕ (11) cos η Z = sin2 ϕ cosθA + cos2 ϕ (12) la = 2asin( η A/2) (13) lb = 2bsin( η Z/2) (14) lt = acos( η A/2) (15) The locations of all vertices can be obtained in any folded state using these parameters. For the pattern shown in Figure 13: Mathematical model of Miura-Ori samples of Fig. 12(a), all straight-crease lines (m) have been the same dimensions and with different density of the numbered beginning from the bottom, and zigzag crease ridges. lines (n) beginning from the left. The vertex at the intersection of the i-th straight crease line and the j-th The Miura-Ori origami structure's mathematical zigzag crease line is denoted as Vi,j, where i = 1, 2,…m, and framework facilitated its digital representation, enabling 1st Author Surname et al.: Paper title 48 AAAA – 2023 – IZOLA - Conference Proceedings us to execute a two-dimensional Fourier transformation. number then decreases to one at the diagonal's apex, The two-dimensional Fourier transform (2D FT) is signifying the lowest spatial frequency. The spatial instrumental in analyzing two-dimensional signals or frequency's averaged outcomes at an individual radius, functions within frequency space. While the primary use scaled to 1/4 wavelength for the 600 x 600 mm sample, of 2D FT is in photo analysis, its applicability extends to across the 2D FT, are conclusively illustrated in Fig. 15. assessing surface geometries, such as surface topography investigations. The operating principal echoes that of the one-dimensional Fourier transform, albeit in two dimensions. For a function, f(x,y), which delineates a surface, the 2D FT is expressed as: ∞ (19) r 𝐹 = (𝑢, 𝑣) = ∬ 𝑓(𝑥, 𝑦)𝑒−𝑖(𝑢𝑥+𝑣𝑦)𝑑𝑥𝑑𝑦 −∞ Here, u and v stand for the spatial frequencies in the x and y directions respectively, and F(u,v) takes the amplitude and phase of the structure at these frequencies. Via spatial frequency space analysis, we can discern various patterns or spatial frequencies on the surface. A surface Figure 14: 2D FT of origami surface magnitude, and with a consistent wave-like pattern, for instance, radius r as descriptor for spatial frequency of diffuse manifests in spatial frequency space as a pronounced sound reflection peak at a specific frequency. The efficacy of the 2D FT lies in contrasting surfaces based on their frequency attributes. It's crucial to understand that in the context of surface geometry analysis, "spatial frequencies" are distinct from temporal frequencies, referencing the rate of surface feature variations across space. Yet, incorporating the speed of sound and sample dimensions, the lowest spatial frequency can be aligned with the sample length's sound wavelength. This alignment allows us to correlate spatial with temporal Figure 15: Correlation between Fourier transformation of frequency. origami surface magnitude and spatial frequency on the The outcome of the 2D FT for surface topography remains radius r for all elements on the two-dimensional Fourier two-dimensional, illustrated in Fig. 14. Given the diffuse transformation surface nature of sound within the reverberation chamber, it's uniformly dispersed around the sample. To transpose the Figure 16 displays the results of the spatial Fourier two-dimensional Fourier transformation into a one- transformation for three distinct Miura-Ori Origami dimensional surface spectrum harmonized to a diffuse samples. While all samples maintain identical dimensions, sound field, the subsequent algorithm was employed: their geometries differ. The visual representation of these samples is provided in Figure 10, with their corresponding 𝑁 1 (20) mathematical models illustrated in Figure 13. The defining 𝐴(𝑟) = ∑ 𝐴(𝑟, 𝑗) 𝑁 parameters of these models are detailed in Table 4. 𝑗=1 The 2D FT distinctly yields spectra characterized by Here, A(r) signifies the mean of all values from the2D FT at prominent peaks. These peaks are representative of the an identical radius r, as showcased in Fig.14. The number density of physical elevations and depressions within the of averages increases with spatial frequency, culminating origami samples. A higher count of these physical features at the spatial frequency's peak for the 1D Fourier. This 1st Author Surname et al.: Paper title 49 AAAA – 2023 – IZOLA - Conference Proceedings in the sample corresponds to a more elevated spatial reverberation times with exceedingly short signal frequency peak in the spectrum. playback durations and the existing criteria for source and background noise level calculations. Leveraging multichannel measurements and new software we enhanced both speed and accuracy of sound absorption measurement. Our investigations validated both hypotheses. Their affirmation ensures the precision of reverberation time measurements within the alpha chamber, setting the stage to effectively evaluate the sound absorption properties of origami. The implemented measuring system demonstrated a high degree of precision, aptly differentiating the acoustic variances across diverse Origami configurations. Its Figure 16: Averaged values of two-dimensional Fourier reliability not only identifies the unique acoustic transformation on the same radius, normalized to 1/4 characteristics of each Origami geometry but also serves wavelength for the sample size of 600 x 600 mm. as a foundation for constructing physical models that describe the acoustic dynamics of Origami structures. Table 5: Comparison of frequencies, for measured sound Incorporating both measurement outcomes and absorption and for maxima of 2D Fourier transformation numerical analyses of the origami surface, there seems to be potential to establish a correlation between the Max. in the Max. in frequencies of sound absorption and the spatial 2D FT [Hz] measured frequencies intrinsic to the origami design, but further absorption [Hz] research is needed. ORIGAMI 2_4, 5, 6 1490 1600 ORIGAMI 1+, 2+, 3+ 1927 1900 5. CONCLUSIONS ORIGAMI 1_1, 2, 3 2974 3150 In this study, we performed a validation of the alpha Extracting the maximal sound absorption frequency chamber and multichannel system for measurements of from measurements and comparing it with the peak reverberation times and consequently for assessment of spatial frequency derived from the sample's mathematical sound absorption, a pivotal metric when evaluating the model reveals a visible correlation. This linkage absorption coefficient of new acoustic materials substantiates the prediction that 1/4 of the sound specifically the Miura-Ori Origami paper structure. Central wavelength can be employed to standardize spatial to this investigation was the development and frequency. Moreover, the radial and planar averaging of deployment of a multichannel software solution tailored the two-dimensional Fourier transformation provides for increased number of measurements reducing the insights into the behavior of the diffuse sound field above influence of non ideal diffuse field. Our findings the sample plane. With this the second hypothesis, that a underscore several relevant points: a significant correlation exists between the sound Software Efficacy: The newly developed software not absorption frequency and the spatial frequency of the only streamlined the process of reverberation time Origami surface, is confirmed. measurements but also enhanced its accuracy. This suggests a promising avenue for its broader adoption in 4. DISCUSSION acoustic research and applications. Influence of Parameters: The study elucidated the Our study has revealed that a minimum of paramount importance of specific parameters in the approximately 50 measurements is essential to obtain measurement process. Specifically, the signal playback valid reverberation times, despite non ideal diffuse field. duration and the total number of measurements We also identified challenges when computing markedly influence the accuracy of reverberation times. 1st Author Surname et al.: Paper title 50 AAAA – 2023 – IZOLA - Conference Proceedings System Resolution: An illustrative test involving plain [5] Ryan L Harne and Danielle T Lynd, Origami acoustics: using thick paper affirmed the system's capability to discern principles of folding structural acoustics for simple and large even minute differences, underscoring its high resolution focusing of sound energy, 2016 Smart Mater. Struct. Vol. 25 pp. and sensitivity. 085031 Convergence to True Value: Our empirical data [6] Pratapa PP, Suryanarayana P, Paulino GH. Bloch wave framework for structures with nonlocal interactions: application to the bolstered the postulate that with an incremental increase design of origami acoustic metamaterials. J Mech Phys Solids in measurement points for the alpha chamber's 2018;118:115–32. reverberation time, we gravitate towards its intrinsic [7] Xiang Yu, Hongbin Fang, Fangsen Cui, Li Cheng, Zhenbo Lu, global value. This convergence phenomenon underscores Origami-inspired foldable sound barrier designs, Journal of the importance of extensive sampling in achieving Sound and Vibration 442 (2019) 514e526 accurate measurements. [8] Diamant R, Shahar I, Rosenberg T, Weiss A. Origami-Inspired Interplay of Sound Wavelength and Room Dimensions: Adaptive Acoustic Tank for Optimal Reflection Mitigation. IEEE A noteworthy revelation from this research was the Sens J 2020;20(24):15193–203. evident correlation between the desired error in global [9] Zou C, Lynd DT, Harne RL. Acoustic wave guiding by reverberation time for the alpha chamber and the ratio of reconfigurable tessellated arrays. Phys Rev Appl sound wavelength to the chamber's dimensions. This 2018;9(1):014009. provides an analytical framework for optimizing [10] Zou C, Harne RL. Tailoring reflected and diffracted wave fields measurements in similar acoustic setups. from tessellated acoustic arrays by origami folding. Wave Motion 2019;89:193–206. Correlation between Sound absorption frequency and [11] Zhu Y, Fei F, Fan S, Cao L, Donda K, Assouar B. Reconfigurable spatial frequency of Origami surface: Presented analysis origami-inspired metamaterials for controllable sound delineated the potential of correlating sound absorption manipulation. Phys Rev Appl 2019;12(3): 034029 frequencies with the spatial frequencies intrinsic to the [12] Tristan Cambonie, Emmanuel Gourdon, Innovative origami-based Miura-Ori origami structure, bridging the gap between solutions for enhanced quarter-wavelength resonators, Journal physical geometry and acoustic behavior. of Sound and Vibration 434 (2018) 379–403 [13] M. Thota, K.W. Wang, Tunable waveguiding in origami phononic In summary, this research not only underscores the structures, Journal of Sound and Vibration 430 (2018) pp.93-100, pivotal role of optimized measurement strategies in [14] ISO 354, Acoustics-Measurement of sound absorption in a acoustic research but also paves the way for novel reverberation room. 2003. methodologies that interlink structural geometry with [15] K. E. Randall and F. L. Ward, “Diffusion of sound in small rooms.” sound behavior, by confirming both hypotheses. [16] A. Cops, J. Vanhaecht, and K. Leppens, “Sound Absorption in a Reverberation Room: Causes of Discrepancies on Measurement Results,” 1995. [17] L. L. Beranek, “Analysis of Sabine and Eyring equations and their 6. REFERENCES application to concert hall audience and chair absorption,” J [1] Larissa M. Fonseca , Guilherme V. Rodrigues , Marcelo A. Savi, An Acoust Soc Am, vol. 120, no. 3, pp. 1399–1410, Sep. 2006, doi: overview of the mechanical description of origami-inspired 10.1121/1.2221392. systems and structures, International Journal of Mechanical [18] T. Vrtovec, “Načrtovanje odmevne komore” Univerza v Ljubljani, Sciences 223 (2022) 107316 Fakulteta za strojništvo, Ljubljana, 2017. [2] Hongbin Fang , Xiang Yu and Li Cheng, Reconfigurable origami [19] Ali Mohammadsalehi, Davood Mostofinejad, Behavior of high-silencers for tunable and programmable sound attenuation, performance concrete canvas Miura-origami structures under Smart Mater. Struct. Vol. 27, (2018), pp. 095007, flexural loading, Structures, Vol. 54, (2023) 928–945 https://doi.org/10.1088/1361-665X/aad0b6 [3] Pingting Jiang, Tianxi Jiang, and Qingbo He, "Origami-based adjustable sound-absorbing metamaterial" Smart Mater. Struct. Vol.30, (2021), pp.057002, https://doi.org/10.1088/1361- 665X/abf420 [4] Xiaomeng Jin, Hongbin Fang, Xiang Yu, Jian Xu, Li Cheng, Reconfigurable origami-inspired window for tunable noise reduction and air ventilationm, Building and Environment 227 (2023) 109802 1st Author Surname et al.: Paper title 51 REVERBERATION TIME ESTIMATION FROM EMOTIONAL SPEECH SIGNALS Andrea Andrijašević Polytechnic of Rijeka, Vukovarska 58, 51 000 Rijeka, Croatia Abstract: Many of today’s speech signal processing systems, from the low-power ones embedded in the hearing aids to the large and complex deep learning based systems for automatic speaker identification, and speech or speech emotion recognition, rely on their speech enhancement pre-processing module to reduce the signal degradation introduced by the room reverberation and background noise. Even though in the former type of systems this type of pre-processing is performed in order to improve the intelligibility of the input speech, whereas in the latter so as to provide feature robustness, in each case the room parameters need to be estimated from the received signal, most commonly the full-band reverberation time (T60). In this study, we examined the robustness of two reverberation time estimation algorithms to changes in speaker’s emotions encoded with the acoustic features such as tempo, voice intensity and pitch height. The results obtained on a set of room impulse responses measured in 11 rooms with T60 in the 0.2 to 1.2 s range, indicate that a spectral decay distributions-based estimator is less sensitive to such type of changes than an estimator based on the modified IS0 3382-2:2008 method. The performance of the latter was successfully enhanced with the introduction of a dynamic range criterion, however it did not surpass the performance of the former algorithm. Keywords: reverberation time, emotion, speech signal, pitch, speech rate, subglottal pressure 1. INTRODUCTION Throughout our lives, we continually experience sometimes quicker, sometimes slower changes in our Speech enhancement is a signal processing procedure conscious mental reactions, subjectively experienced as commonly performed in a wide array of electronic strong feelings and typically accompanied by physiological consumer devices with integrated microphones, e.g. and behavioural changes, which we commonly refer to as mobile phones, tablets, and hearing aids, with the aim of emotions [9]. In speech communication, many acoustic reducing the damaging effects of background noise and features, such as pitch, sound intensity, and tempo carry reverberation either on the performance of systems for important emotional information [10]. As the changes in automatic speech, speaker or emotion recognition or on emotional state can be sudden, it is even more important the intelligibility of received speech [1-5]. For this task, an to evaluate the robustness of algorithms to them. It is well estimate of room reverberation time ( T 60) is often known that the ratio of the duration of vowels to the required, and thus usually obtained from the received duration of consonants changes with speech rate [11], reverberant speech signal itself. that the subglottal pressure affects the tilt of the spectral Even though both the influence of background noise and envelope and the overall sound pressure level (SPL), and speech phonetic content on the performance of that speaker’s intonation is reflected in the fundamental algorithms for T 60 estimation have been well researched frequency (F0) of voiced speech sounds [12]. [6-8], the influence of speaker’s emotional state remains Therefore, in this study we assess the robustness of T 60 unexplored since the speech material used in [6-8] was estimation algorithms to emotions conveyed through uttered in only one speaking style. changes in speaker’s fundamental frequency, speech rate and subglottal pressure. To this end, we use phonetically Andrijašević: Reverberation time estimation from emotional speech signals 52 AAAA – 2023 – IZOLA - Conference Proceedings balanced speech material uttered in six different speaking 3. SPEECH RECORDINGS styles convolved with room impulse responses measured in 11 rooms at up to 5 different source-receiver distances, The OLdenburg LOgatome (OLLO) corpus, a speech from which T 60 is estimated using two of the best database designed for the comparison of recognition performing algorithms from [7]. performance of automatic speech recognizers and human The remainder of the paper is organised as follows. In listeners, ver. 2.0 [19] was used in this study. Namely, we Section 2, we present the algorithms used for T 60 used the recordings of 50 phonetically balanced sentences estimation, whereas in Sections 3 and 4 we provide the in French language, uttered in six speaking styles by nine analyses of emotional speech recordings and room speakers whose identification numbers are as follows: impulse responses, respectively. We continue by o female speakers: 41, 45, 47, and 49, presenting the results in Section 5 and proposing two o male speakers: 42, 43, 44, 46, and 48. algorithm modifications in Section 6. Finally, with Section This set was chosen as it consists of recordings of the same 7, we conclude our work. speech material uttered using different tempi (speech rates), different intonations (pitch curves), and in which different SNRs arise due to changes in the subglottal 2. T60 ESTIMATION ALGORITHMS pressure the speaker uses, which cannot be imitated by adding different levels of pre-recorded or synthetic noise The first algorithm for T 60 estimation used in this study is to the normal style speech recordings. In the following Prego’s algorithm [13, 14], in which estimation is carried subsections, we present and analyse the aforementioned out in the short-time Fourier transform (STFT) domain. aspects of the recordings. The so-called free-decay regions of speech signal are detected independently in each of the frequency bands in 3.1. Signal-to-noise ratio the 0 - 4 kHz range. Estimation of T 60 is performed on the energy decay curves (EDCs) calculated from these free-Noise power was estimated from the first 100 ms of the decays using a procedure that closely resembles recordings, which contained no speech activity. Schroeder’s method [15, 16]. The final estimate is Interestingly, it was observed that the noise power obtained as the output of a linear mapping function spectral density (PSD) differed between speakers. This whose input is the median of frequency bands T 60 was due to breathing noise – the recording microphone medians. had been placed about 25 cm from the speaker’s mouth, The second algorithm, Eaton’s, operates in the mel- hence the inspiration and expiration noise was quite frequency STFT domain [17]. In this algorithm, the sound noticeable for a few speakers whereas others succeeded decay rates are estimated for each mel-frequency band in directing minimal noise towards it. With most of its independently as the slope of a linear least-squares fit. power in the 60 - 600 Hz range [20, 21], breathing noise The negative-side variance (NSV), defined as the variance became superimposed onto recording equipment noise. of negative slopes in the distribution of decay rates, is Speech signal power was calculated as the active speech used as the input to a polynomial mapping function whose level of the recording, estimated using the ITU-T P.56 output is the estimate of T 60. Method B [22, 23], from which the previously estimated The performance of these two algorithms under different noise power was subtracted. For the normal style speech background noise conditions and with different phonetic recordings, the median SNR across the speakers was in the content has been well researched and documented in [7, 39 - 55 dB range. 8, 18], enabling us to focus on the evaluation of their As the initial tests had shown that a single sentence robustness to emotional speech signals. Accordingly, since recording was not sufficiently long to produce a T 60 high signal-to-noise ratio (SNR) recordings were used in estimate, the 50 recordings uttered by a single speaker in this study, no signal de-noising was performed and the one speaking style needed to be concatenated into one linear mapping function parameters were set as in [13] for speech file. To this end, since the OLLO corpus recordings Prego’s algorithm, whereas in Eaton’s algorithm the NSV come normalized, in order to make the background noise was calculated from the distribution of all obtained of the concatenated recordings constant, the individual negative decay rates. Andrijašević: Reverberation time estimation from emotional speech signals 53 AAAA – 2023 – IZOLA - Conference Proceedings recordings were first de-normalized using the envelope of voiced speech sounds spectra, where the normalization factors provided with the corpus. steeper the slope, the lower the SGP. Figure 2 presents the long-term average spectra of the 54 3.2. Speech rate root-mean-square (RMS) normalized speech files, averaged across the same sex speakers files. Changes in Since the same phonetically balanced speech material was the underlying subglottal pressure are visible in the 1 - 5 uttered in six speaking styles, the ratio of the fast, slow, kHz range – LTAS, and therefore the SGP as well, has loud, soft, and questioning speaking styles speech considerably higher values for the loud style and duration to the duration of the normal style speech somewhat lower values for the soft style than for the informs of the change in the rate of speech introduced by remaining four speaking styles. the speaker. Therefore, for each of the 54 speech files, a speech duration estimate was calculated by multiplying sex: F sex: M 40 40 the speech file nominal duration in seconds by its activity Fast 35 35 Slow factor, defined as the ratio of the voice active to the total 30 30 Loud number of speech samples in a file, obtained with [23]. Soft 25 25 Questioning Figure 1 shows the calculated speech duration ratios. They B) 20 B) 20 (d (d Normal agree quite well between the sexes. Interestingly to AS 15 AS T 15 T L L notice, the speech rate for the questioning style is 10 10 somewhat higher than for the normal style. An outlier, a 5 5 data point with the value of 2.05, present for the slow 0 0 style, belongs to the female speaker number 49. -5 -5 0 2 4 6 8 0 2 4 6 8 Frequency (kHz) Frequency (kHz) sex: F sex: M 2 2 Fig. 2. LTAS of the RMS-normalized speech files 1.8 1.8 tio tio ra 1.6 ra 1.6 n n tio tio 3.4. Fundamental frequency ra 1.4 ra u 1.4 u d d ch ch e 1.2 e 1.2 e e Figures 3 and 4 show the ratios of the estimates of Sp Sp 1 1 fundamental frequency mean and standard deviation obtained in Praat [24] for single sentence speech 0.8 0.8 recordings, with normal style as reference. F Sl L So Q F Sl L So Q The mean F0 ratios change across the five speaking styles Fig. 1. Speech duration ratios, with normal speaking style similarly for female and male speakers, with mean F0 as reference. x-axis labels: ‘F’ – fast, ‘Sl’ – slow, ‘L’ – loud, being about 1.3 times higher for the loud and questioning ‘So’ – soft, and ‘Q’ – questioning speaking style styles than for the normal style. Although F0 is controlled primarily by laryngeal muscle activity, the subglottal 3.3. Subglottal pressure pressure also has an influence on its value – as the SGP increases, the F0 increases as well [12], which explains the In controlled settings, subglottal pressure (SGP) can be increase observed for the loud style. tracked by analysing the flow glottogram, whose peak Figure 4 shows that the changes in the ratio of the flow and the negative peak amplitude of its derivative, standard deviation of F0 across the five speaking styles are referred to as the maximum flow declination rate, also quite similar between the sexes. Since, among its increase with SGP [12]. In case a glottogram is not many functions, F0 is used to signal syntactic information available, the changes in the subglottal pressure, such as the difference between statements (normal style) perceived as changes in loudness, can be (under the and questions (questioning style), where the latter are assumption of a fixed speaker-microphone distance) recognized by a sharp rise in its value as the speaker estimated either from the changes in the active speech approaches the end of the final word in a sentence [25], level (ASL) of the recording or from the slope of the Andrijašević: Reverberation time estimation from emotional speech signals 54 AAAA – 2023 – IZOLA - Conference Proceedings an increase in its variation is present for the questioning their mean ground truth full-band T 60 (from here onwards style recordings (Figure 4) in addition to the previously denoted as T GT) [7], are presented in Table 1. observed increase in its mean value (Figure 3). The AIR set comprises impulse responses measured in four rooms. In each room, RIRs were measured using two sex: F sex: M 1.8 1.8 microphones and, depending on the room size, either three or five different source-receiver distances were 1.6 1.6 utilized. For each room impulse response of this set, the oi o t it associated T GT was calculated in accordance with [15], as 1.4 ra 1.4 ra ) ) 0 0 the T 20 obtained from its full-band energy decay curve. (F (F n n a 1.2 a 1.2 The mean T GTs are presented in Table 2. The variance of Me Me T GT was less than 4.4∙10-3 s2 in the rooms, thus validating 1 1 the assumption of position-invariant reverberation time. As can be seen from Tables 1 and 2, the ACE and AIR sets 0.8 0.8 F Sl L So Q F Sl L So Q jointly cover a large T GT range, which coincides with the ranges reported in [27] and [28] for private, public, and Fig. 3. Ratios of mean F0, with normal speaking style as work spaces. reference Room DRR DRR min max TGT (s) SFmin SFmax sex: F sex: M name (dB) (dB) 4 4 Office 1 0.332 -0.61 8.85 0.82 0.87 3.5 3.5 Meeting 3 3 0.371 3.94 7.96 0.83 0.87 room 2 oi o t 2.5 it 2.5 ra ra ) ) Office 2 0.390 1.80 9.74 0.81 0.87 0 2 0 2 (F (F d d St 1.5 St 1.5 Meeting 0.437 3.21 7.89 0.83 0.88 1 1 room 1 0.5 0.5 Lecture 0.638 1.03 8.04 0.83 0.91 0 0 room 1 F Sl L So Q F Sl L So Q Building Fig. 4. Ratios of standard deviation of F0, with normal 0.646 1.46 8.52 0.83 0.86 lobby speaking style as reference Lecture 1.220 -0.33 6.00 0.84 0.90 room 2 Table 1. The ACE set 4. ROOM IMPULSE RESPONSES Two sets of measured room impulse responses (RIRs) were utilized in this study. The first set consists of 28 Room DRRmin DRRmax TGT (s) SFmin SFmax impulse responses from the ACE Corpus [7], whereas the name (dB) (dB) second set comprises 32 impulse responses from the Booth 0.178 1.55 10.01 0.83 0.92 Aachen Impulse Response (AIR) database v1.4 [26]. Meeting 0.277 2.10 3.93 0.81 0.84 room 4.1. Ground truth T60 Office 0.546 -3.19 5.07 0.80 0.90 Lecture The ACE room impulse responses were obtained with a 0.819 -7.52 3.01 0.80 0.84 room linear eight-channel microphone array, namely its channels 1 and 8, at two different source-receiver Table 2. The AIR set distances within seven different rooms. The estimates of Andrijašević: Reverberation time estimation from emotional speech signals 55 AAAA – 2023 – IZOLA - Conference Proceedings 4.2. Direct-to-reverberant ratio for each T GT, the source-receiver distance increases from left to right, by which neither algorithm is substantially As the room impulse responses were measured at up to affected. Figures also indicate that the largest ISDs occur five different source-receiver distances, their direct-to- for the loud and soft speaking styles, as well as that, as the reverberant ratio (DRR) was calculated using the method T GT increases, the absolute value of ISD decreases. For presented in [7]. The minimal and maximal DRRs shown in Prego’s algorithm, a similar, but less pronounced, trend as Tables 1 and 2 indicate that the majority of RIRs were for the loud style can be observed for the questioning measured at distances smaller than the critical distance style. [1], at which the energy of the direct path component is Based on the results presented in Figures 5 and 6, it can higher than the total energy of reflections. be concluded that even though Prego’s algorithm performed somewhat better than Eaton’s in [7], the latter 4.3. Spectral flatness is, at high SNRs, robust to a number of changes that speakers introduce, with more than 95 % of ISDs within In case the magnitude spectrum of a room impulse the [-50, 50] ms range. Since Eaton’s algorithm only response is flat, the SNR of a clean speech recording can switches the decay rate sub-selection rule based on the be used as a proxy for the SNR of its reverberant version. estimated SNR, it is not unreasonable to assume that the To check whether this applies to the ACE and AIR RIRs, with the introduction of a de-noising step its performance spectral flatness (SF) [29], the ratio of the geometric to the under low SNR conditions could be further improved. arithmetic mean of the magnitude spectrum coefficients, was calculated in the 0 - 8 kHz range. The minimal and 5.2. Prego’s algorithm – detailed results maximal SFs given in Tables 1 and 2 indicate that the magnitude spectra are indeed flat, thus enabling the use In order to find the cause for the large ISDs observed for of clean speech SNRs in the analyses presented in the Prego’s algorithm, we explore their relationship with following sections. speech signal characteristics. Since the largest ISDs are present for the loud and soft styles, we focus on the relationship of ISD with SGP. In the following analysis, the 5. RESULTS changes in the ASL of clean speech files are used as a proxy 5.1. Overall results for the changes in the utilized SGP under the previously verified assumptions of spectral flatness and noise power The performance of the algorithms was first assessed with invariability across the clean speech files of different the mean squared error and estimation bias, i.e., the speaking styles. mean error, for the T 60 estimates obtained from the Given this, the relationship between the ASL ratio and ISD normal style speech files as they are later used as baseline is presented in Figure 7, with the former defined as the in the analyses of inter-style differences. The first ratio of the ASL of the first five styles to the ASL of the algorithm, Prego’s, achieved a mean squared error of normal style. As can be seen from Figure 7, there is a 0.095 s2 and a bias of 0.269 s, whereas Eaton’s performed considerable variation in how much the speakers changed better with a mean squared error of 0.017 s2 and a bias of the SGP across the speaking styles. Moreover, it can be 0.058 s. also observed that some of the speakers increased the Figures 5 and 6 present the inter-style differences (ISDs), loudness of their voice for the questioning style as well, defined as the difference between the T 60 estimates which had in turn made their questioning style ISDs more obtained from the fast, slow, loud, soft and questioning similar to those present for the loud style. style speech files and the corresponding T 60 estimate Finally, it can be summarized: obtained from the normal style speech file. It should be o as the SGP increases above (decreases below) its noted that since the Pearson correlation coefficient normal level, the ISD becomes negative (positive), revealed that for both algorithms the T 60 estimates o the larger the change in SGP, the higher the absolute relationship with T GT is linear (PCC higher than 0.9), the value of ISD, bias observed for Prego’s algorithm was not corrected as o for longer T GT (> 0.6 s), the influence of SGP on ISD it had no influence on the ISD curve shape. In both Figures, becomes considerably reduced. Andrijašević: Reverberation time estimation from emotional speech signals 56 0.3 Fast Slow Loud Soft Questioning 0.3 0.2 Fast Slow Loud Soft Questioning (s) 0.2 ce 0.1 (s)nrece 0.1 niffe 0 re deiffyl 0 d -0.1 er-stylte -0.1 r-stIn -0.2 teIn -0.2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 8 7 2 1 7 6 8 6 9 8 7 2 1 7 6 8 6 9 8 7 2 1 7 6 8 6 9 8 7 2 1 7 6 8 6 9 8 7 2 1 7 6 8 6 9 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 80 70 20 10 0 70 60 80 60 90 1 80 70 20 10 0 70 60 80 60 90 1 80 70 20 10 0 70 60 80 60 90 1 80 70 20 10 0 70 60 80 60 90 1 80 70 20 10 0 70 60 80 60 90 1 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 1 Fig. 5. Inter 2 -style differences for 3 Prego’s algorithm. T 4 5 GT is shown on the x-axis 1 2 3 4 5 0.3 0.3 Fast Slow Loud Soft Questioning 0.2 Fast Slow Loud Soft Questioning (s) 0.2 cen (s) 0.1 rece 0.1 niffere 0 deiffe 0 yl de -0.1 r-stylte -0.1 Inr-stte -0.2 In -0.2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 1 2 3 1 2 3 4 5 1 2 1 2 1 2 1 2 1 2 3 1 2 1 2 1 2 3 4 5 1 2 8 7 2 1 7 6 8 6 9 8 7 2 1 7 6 8 6 9 8 7 2 1 7 6 8 6 9 8 7 2 1 7 6 8 6 9 8 7 2 1 7 6 8 6 9 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 .18 .27 .32 .31 .3 .47 .56 .68 .66 .89 .2 .18 .27 .32 .31 .3 .47 .56 .68 .66 .89 .2 .18 .27 .32 .31 .3 .47 .56 .68 .66 .89 .2 .18 .27 .32 .31 .3 .47 .56 .68 .66 .89 .2 .18 .27 .32 .31 .3 .47 .56 .68 .66 .89 .2 07 07 03 07 09 03 04 03 04 01 12 07 07 03 07 09 03 04 03 04 01 12 07 07 03 07 09 03 04 03 04 01 12 07 07 03 07 09 03 04 03 04 01 12 07 07 03 07 09 03 04 03 04 01 12 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 1 Fig. 6. Inter 2 -style differences for 3 Eaton’s algorithm. T GT4 is shown on the x-axis5 1 2 3 4 5 Speaker: 41, sex: F Speaker: 42, sex: M Speaker: 43, sex: M Speaker: 44, sex: M Speaker: 45, sex: F 0.6 0.6 0.6 0.6 0.6 0.4 0.4 0.4 0.4 0.4 (s) (s) (s) (s) (s) ce ce ce ce ce n 0.2 n 0.2 n 0.2 n 0.2 n 0.2 re re re re re iffe 0 iffe 0 iffe 0 iffe 0 iffe 0 d d d d d e e e e e yl -0.2 yl -0.2 yl -0.2 yl -0.2 yl -0.2 r-st r-st r-st r-st r-st te -0.4 te te te te In -0.4 In -0.4 In -0.4 In -0.4 In -10 0 10 20 -10 0 10 20 -10 0 10 20 -10 0 10 20 -10 0 10 20 Active speech level ratio (dB) Active speech level ratio (dB) Active speech level ratio (dB) Active speech level ratio (dB) Active speech level ratio (dB) Speaker: 46, sex: M Speaker: 47, sex: F Speaker: 48, sex: M Speaker: 49, sex: F 0.6 0.6 0.6 0.6 0.4 0.4 0.4 0.4 Fast (s) (s) (s) (s) ce ce ce ce Slow n 0.2 n 0.2 n 0.2 n 0.2 re re re re Loud iffe 0 iffe 0 iffe 0 iffe 0 Soft d d d d e e e e yl Questioning -0.2 yl -0.2 yl -0.2 yl -0.2 r-st r-st r-st r-st te -0.4 te te te In -0.4 In -0.4 In -0.4 In -10 0 10 20 -10 0 10 20 -10 0 10 20 -10 0 10 20 Active speech level ratio (dB) Active speech level ratio (dB) Active speech level ratio (dB) Active speech level ratio (dB) Fig. 7. Relationship between the ASL ratio and ISD for Prego’s algorithm. Open circles: RIRs with T GT < 0.6 s, Filled diamonds: RIRs with T GT > 0.6 s Andrijašević: Reverberation time estimation from emotional speech signals 57 AAAA – 2023 – IZOLA - Conference Proceedings 100 100 Fast Slow Loud Soft Questioning Normal ) ) (% 80 80 (% B B d d 5 5 2 2 = = > 60 > 60 e e g g n n ra ra F B B d 40 d 40 Sl a a ith ith L w w s s So C 20 C 20 D D E E Q N 0 0 30 40 50 60 70 8 7 2 1 7 6 8 6 9 8 7 2 1 7 6 8 6 9 8 7 2 1 7 6 8 6 9 8 7 2 1 7 6 8 6 9 8 7 2 1 7 6 8 6 9 8 7 2 1 7 6 8 6 9 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 7 7 3 7 9 3 4 3 4 1 2 median SNR (dB) .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 .1 .2 .3 .3 .3 .4 .5 .6 .6 .8 .2 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 1 2 3 (a) 4 5 6 (b) Fig. 8. a) Percentage of EDCs with a dynamic range equal to or greater than 25 dB. b) Relationship between the percentage of EDCs with a dynamic range equal to or greater than 25 dB for Office 2 RIRs and the SNR of clean speech recordings 6. ISO 3382-BASED MODIFICATIONS with the Office 2 RIRs, the percentage of the curves that satisfied the linearity criterion was 30 % for the loud, 25 % The results presented in Section 5 clearly show that for the normal, and 23 % for the soft style. Prego’s algorithm is primarily sensitive to the changes in Hence, in this modification, all the sub-band estimates of the subglottal pressure – a well-known sensitivity that T 60 obtained from an energy decay curve with ξ above 10 Schroeder’s method exhibits to changes in the ratio of the ‰ were discarded, upon which the standard procedure, room sound excitation to background noise level [15] i.e. calculation of the final T 60 estimate as the median of reappeared, this time in the STFT domain. sub-band median T 60s, was performed. Unfortunately, no In order to try to alleviate the observed influence of SGP improvement could be observed – the ISDs were similar on algorithm’s performance, two different modifications to the ones obtained with the original algorithm. based on two ISO 3382-2 measures of the quality of energy decay curves [15] are proposed. In both of the 6.2. Dynamic range modified versions of the algorithm, the first part of it was left unchanged, i.e. the original number of frequency In this modification, the dynamic range of each sub-band bands as well as the procedure for the detection of sub- energy decay curve used for T 60 estimation was calculated, band free-decay regions and subsequent T 60 estimation and a criterion of 25 dB, a value half way between the two were used, so that the same sub-band T 60s were obtained standard T 60 estimation ranges [15], was set. Panel a) in as with the original algorithm. These sub-band T 60 Figure 8 shows the percentage of the energy decay curves estimates are then processed in one of the two following of a reverberant speech file with a dynamic range equal to ways. or greater than 25 dB that, not surprisingly, depends on the speaking style. To illustrate further how the 6.1. Degree of non-linearity percentage of EDCs with a dynamic range greater than or equal to 25 dB relates to the SNR of clean speech In the first ISO 3382-based modification, the degree of recordings, the results for the Office 2 RIRs are presented non-linearity ( ξ), a measure obtained from the correlation in Panel b). coefficient of the linear least-squares fit to the energy Furthermore, in Panel a), for all speaking styles, the decay curve [15], was calculated for each detected speech percentage of EDCs with a high dynamic range decreases free-decay region. It was observed that a very large as the T GT increases due to temporal smearing of the proportion of decay curves had ξ higher than the speech signal that the room impulse response introduces maximum of 10 ‰ allowed for Schroeder’s method by the – the higher the T GT, the more of the fine structure of the Standard. For instance, when speech files were convolved speech signal envelope is lost. As a consequence of this Andrijašević: Reverberation time estimation from emotional speech signals 58 AAAA – 2023 – IZOLA - Conference Proceedings increasingly more severe loss of modulation depth, the 6.3. Discussion differences in the percentages across the speaking styles are largest for the smallest values of T GT and slowly After exploring the results of two modifications of Prego’s disappear as the T GT increases. algorithm, where neither of them could reduce the ISDs to After all the sub-band T 60 estimates obtained from EDCs of the level observed for Eaton’s algorithm, it seems likely dynamic range lower than 25 dB were discarded, and the that the robustness that the latter exhibits comes from same standard procedure, i.e. calculation of the final T 60 the use of wide mel-bands and processing of the whole estimate as the median of sub-band median T 60s, was reverberant signal. Due to its use of mel-bands, Eaton’s performed, an improvement in algorithm robustness was algorithm places more emphasis to T 60 estimation in the obtained for T GT lower than 0.6 s. There was a significant lower frequency region where most of vowel energy is increase in the percentage of ISDs within the [-50, 50] ms contained, whereas Prego’s algorithm distributes weight range – from 41 % for the original algorithm to almost 81 equally across the frequency bands in the 0 - 4 kHz range. % for the modified version. Unfortunately, for T GT above In addition to that, in Eaton’s algorithm each sub-band 0.6 s the percentage of ISDs within that range decreased speech decay can potentially produce multiple negative from 75 % for the original algorithm to 48 % for the gradients, whereas in Prego’s algorithm one detected sub- modified version. band free-decay always equals one sub-band T 60 estimate. Given this improvement in algorithm’s robustness to changes in speaking style for T GT lower than 0.6 s, the performance metrics for the normal style speech files 7. CONCLUSION were re-calculated. The following results were obtained – a decrease in the value of correlation coefficient to 0.76, The artificially reverberated phonetically balanced speech a mean squared error of 0.034 s2, and a bias of -0.048 s. material uttered in six different speaking styles proved to The decrease in the correlation coefficient occurred due be a challenging test set for the task of T 60 estimation. to estimates obtained from the room with the highest T GT Specifically, it was shown that even in high SNR conditions (Lecture room 2). After the data for that room were a T 60 estimator based on Schroeder’s method remains removed, a linear trend across the T GTs and a decrease in sensitive to changes in the subglottal pressure. With an the estimation bias were revealed – a correlation ISO 3382-based dynamic range criterion for sub-selection coefficient of 0.94 for the modified versus 0.96 for the of free-decay regions, a reduction in its sensitivity was original algorithm, a mean squared error of 0.005 s2 versus achieved for rooms with T 60 lower than 0.6 s, yet it 0.099 s2, and a bias of -0.004 s for the modified versus remained higher than for an approach of decay rate 0.273 s for the original algorithm. estimation in mel-frequency bands. Based on these results and the data presented in Figure 8, Finally, since robust estimation of T 60 from emotional it seems that, even though the silent parts of the speech speech signals remains a challenge, we strongly files between individual sentences were of sufficient encourage wider use the OLLO corpus in future duration for the estimation of T GT larger than 0.6 s (they performance evaluation studies alongside other, well-were about 1 second long), the value of 0.6 s can be known speech corpora in which the speech material is considered a threshold above which a dynamic range uttered in only one speaking style. criterion is not useful when trying to achieve robustness to changes in speaking style. Finally, given the increasingly smaller percentage of EDCs with a dynamic range above ACKNOWLEDGMENT 25 dB as the T GT increases, it also seems that, in practice, The author would like to thank T. Prego for generously the estimation of T 60 on connected speech in the STFT providing his algorithm and J. Eaton for making his domain can be performed successfully only for a subset of algorithm publicly available. The author would also like to T GT values, above whose upper limit it would be more express her gratitude to P. A. Naylor from Imperial College appropriate to estimate the early decay time ( T 10). London for kindly providing the ACE Corpus room impulse responses. Andrijašević: Reverberation time estimation from emotional speech signals 59 AAAA – 2023 – IZOLA - Conference Proceedings 8. REFERENCES and direct-to-reverberant energy ratio using subband speech decomposition, in Proc. IEEE [1.] Naylor, P. A. and Gaubitch, N. D. Speech Workshop on Applications of Signal Processing to Dereverberation. Springer, 2010. Audio and Acoustics (WASPAA), 2015. [2.] Tashev, I. J. De-reverberation. In Sound Capture and [15.] ISO 3382-2:2008. Acoustics - Measurement of Room Processing: Practical approaches. Wiley, pp. 341- Acoustic Parameters - Part 2: Reverberation Time in 358, 2009. Ordinary Rooms, Intl. Org. for Standardization, 2009. [3.] Löllmann, H. W. and Vary, P. Low delay noise [16.] Schroeder, M. R. New method of measuring reduction and dereverberation for hearing aids, reverberation time, J. Acoust. Soc. Am., 37(3), pp. EURASIP Journal on Advances in Signal Processing, pp. 409–412, 1965. 1-9, 2009. [17.] Eaton, J., Gaubitch, N. D. and Naylor, P. A. Noise- [4.] Kinoshita, K., Delcroix, M., Gannot, S., Habets, E. A. P., robust reverberation time estimation using spectral Haeb-Umbach, R., Kellermann, W., Leutnant, V., decay distributions with reduced computational Mass, R., Nakatani, T., Raj, B., Sehr, A. and Yoshioka, cost, in Proc. IEEE Intl. Conf. on Acoustics, Speech and T. A summary of the REVERB challenge: state-of-the- Signal Processing (ICASSP), pp. 161-165, 2013. art and remaining challenges in reverberant speech [18.] Eaton, J., Gaubitch, N. D., Moore, A. H. and Naylor, P. processing research, EURASIP Journal on Advances in A. ACE Challenge results technical report, Imperial Signal Processing, 2016(7), pp. 1-19, 2016. College London, 2016. [5.] Lee, C.-C., Sridhar, K., Li, J.-L., Lin, W.-C., Su, B.-H. and [19.] Wesker, T., Meyer, B., Wagener, K., Anemüller, J., Busso, C. Deep Representation Learning for Affective Mertins, A. and Kollmeier, B. Oldenburg Logatome Speech Signal Analysis and Processing: Preventing Speech Corpus (OLLO) for Speech Recognition unwanted signal disparities, IEEE Signal Processing Experiments with Humans and Machines, in Magazine, 38(6), pp. 22-38, 2021. Proceedings of Interspeech, pp. 1273-1276, 2005. [6.] Gaubitch, N. D., Löllmann, H. W., Jeub, M., Falk, T. H., [20.] Gross, V., Dittmar, A., Penzel, T., Schüttler, F. and von Naylor, P. A., Vary, P. and Brookes, M. Performance Wichert, P. The relationship between normal lung comparison of algorithms for blind reverberation sounds, age and gender, Am. J. Respir. Crit. Care time estimation from speech, in Proc. Intl. Workshop Med., 162, pp. 905-909, 2000. on Acoustic Signal Enhancement (IWAENC), 2012. [21.] Niu, J., Cai, M., Shi, Y., Ren, S., Xu, W., Gao, W., Luo, [7.] Eaton, J., Gaubitch, N. D., Moore, A. H. and Naylor, P. Z. and Reinhardt, J. M. A novel method for automatic A. Estimation of room acoustic parameters: The ACE identification of breathing state, Nature Scientific Challenge, IEEE/ACM Trans. Audio, Speech, and Reports, pp. 1-13, 2019. Language Process., 24(10), pp. 1681-1693, 2016. [22.] Objective Measurement of Active Speech Level, [8.] Andrijašević, A. Effect of phoneme variations on blind International Telecommunications Union (ITU-T), reverberation time estimation, Acta Acustica, 4(1), Recommendation P.56, 1993. pp. 1-17, 2020. [23.] Brookes, D. M. VOICEBOX: A speech processing [9.] Emotion - Merriam-Webster dictionary. Available at: toolbox for MATLAB. Available at: https://www.merriam- http://www.ee.ic.ac.uk/hp/staff/dmb/voicebox/voic webster.com/dictionary/emotion [10.] Quinto, L., Thompson, W. F. and Keating, F. L. ebox.html Emotional communication in speech and music: the [24.] Boersma, P. and Weenink, D. Praat: doing phonetics role of melodic and rhythmic contrasts, Front. by computer, version 5.1.43. University of Psychol., 4(184), 2013. Amsterdam, 2010. Available at: [11.] Kuwabara, H. Acoustic properties of phonemes in https://www.fon.hum.uva.nl/praat/ continuous speech for different speaking rate, in [25.] Rose, P. The human vocal tract and the production Proc. 4th International Conference on Spoken and description of speech sounds. In Forensic Language, pp. 2435-2438, 1996. Speaker Identification. Taylor & Francis, pp. 125-174, [12.] Rossing, T. D. The human voice in speech and 2002. singing. In Springer Handbook of Acoustics. Springer, [26.] Jeub, M., Schäfer, M. and Vary, P. A binaural room pp. 669-712, 2007. impulse response database for the evaluation of [13.] Prego, T. M., Lima, A. A. and Netto, S. L. A blind dereverberation algorithms, in Proc. Intl. Conf. on algorithm for reverberation-time estimation using Digital Signal Processing, pp. 1-4, 2009. subband decomposition of speech signals, J. Acoust. [27.] Diaz, C. and Pedrero, A. The reverberation time of Soc. Am., 131(4), pp. 2811–2816, 2012. furnished rooms in dwellings, Applied Acoustics, 66, [14.] Prego, T. M., Lima, A. A., Zambrano-Lopez, R. and pp. 945-956, 2005. Netto, S. L. Blind estimators for reverberation time Andrijašević: Reverberation time estimation from emotional speech signals 60 AAAA – 2023 – IZOLA - Conference Proceedings [28.] Vincent, E., Gannot, S. and Virtanen, T. Acoustics – [29.] Johnston, J. D. Transform coding of audio signals Spatial properties. In Audio Source Separation and using perceptual noise criteria, IEEE Journal on Speech Enhancement, Wiley, pp. 31-46, 2018. Selected Areas in Communications, 6(2), pp. 314-323, 1988. Andrijašević: Reverberation time estimation from emotional speech signals 61 DETERMINATION OF THE POSITION OF EQUIPMENT NOISE SOURCES IN EDUCATIONAL INSTITUTIONS ACCORDING TO SUBJECTIVELY EVALUATED SPEECH INTELLIGIBILITY Mateja Dovjak 1*, Denis Pirnat 2, Rok Prislan 3 1,2 Univerza v Ljubljani, Fakulteta za Gradbeništvo in Geodezijo, Jamova 2, 1000 Ljubljana, Slovenija, mateja.dovjak@fgg.uni-lj.si 3 InnoRenew CoE, Livade 6a, SI-6310 Izola, Slovenia, rok.prislan@innorenew.eu Abstract: Building-equipment noise is the primary source of background noise in buildings. It is a key factor influencing speech intelligibility, along with room acoustics, sound insulation of the building envelope and the noise level in the room. The impact of building-equipment noise on the acoustic environment in educational institutions is poorly addressed in current design practice; attention is focused only on the requirements for maximum background noise levels and not on the position of noise sources in the room. Our study focuses on a characteristic lecture hall of an educational institution where an artificial speaker was used to simulate the speaker, while the noise typically produced by an HVAC system was generated by an omnidirectional loudspeaker. The position and level of the generated noise were varied for two listening/microphone positions. Under these conditions, (i) the Speech Transmission Index (STI), which objectively determines speech intelligibility, was measured, (ii) listening tests were performed to assess intelligibility of digits, and (iii) subjective assessment of intelligibility was performed using questionnaires. On this basis, recommendations for speech intelligibility in lecture halls and classrooms were developed in terms of equipment position and background noise level. The results show that although the position of the noise source has no significant influence on the STI, it can influence the perceived intelligibility. As such, background noise is not a sufficient criterion for including noise sources in the built environment, and the position of noise sources can be optimised in the design phase. This research was approved by the Ethics Committee for Research Involving Human Subjects, University of Ljubljana (012-2021; Ljubljana, 18 February 2022). Keywords: equipment noise, speech intelligibility, noise source position 1. INTRODUCTION to adapt to background noise); ii) listening ability (hearing disabilities, native language); and iii) external factors The impact of background noise, particularly building- related to the room (room acoustic properties, level of equipment noise, is an important aspect to consider in the background noise). As this way of delivering information building construction process to achieve a high quality is the basis of today's educational system, it is crucial to acoustic environment. Studies [1-3] show that poor room minimise external influences on speech intelligibility when acoustics, insufficient sound insulation of the building planning classrooms and lecture halls, as designers have envelope and elevated noise levels in the room due to no influence on speaking or listening abilities. To ensure numerous technical building systems installed for building optimal speech intelligibility in educational institutions, it operation, often negatively affect speech intelligibility. is therefore necessary to focus on the acoustic properties Speech intelligibility plays an essential role in the of the room, such as optimal reverberation time and successful speech delivery. It depends on i) speaker effective sound insulation of the external and internal speaking ability (level and intelligibility of speech, ability structural assemblies, the reduction of noise levels in the Dovjak et al.: Determination of the position of equipment noise sources in educational institutions according to the subjectively evaluated speech intelligibility 62 AAAA – 2023 – IZOLA - Conference Proceedings lecture hall and the appropriate positioning of noise protection in buildings [7] specifies the maximum sources [4]. acceptable sound pressure levels LAeq = 30 dB(A), and noise generate by equipment LAFmax = 40 dB(A). The impact of noise from technical building systems on the acoustic environment in educational institutions was This study aims to investigate the influence of background studied by Serpilli et al. [2]. They investigated the impact noise levels and the position of the ventilation system of mechanical ventilation systems on the noise level and (HVAC) on the speech intelligibility in the lecture hall. Our speech transmission in a lecture hall, which is determined research objectives are: i) to record speech produced by using the STI parameter. They found that the introduction an artificial speaker at different listening/microphone of mechanical ventilation increases the ambient noise position in the room, where an additional noise source levels and decreases the STI values, which proves to be (positioned at different locations in the room and particularly problematic in rooms that are already reproducing noise at different levels) generates noise as acoustically inadequate (in the study, rooms with too long would be produced by the HVAC, ii) to measure the reverberation time). They additionally point out the poor Speech Transmission Index (STI), which objectively installation of the mechanical ventilation system: determines speech intelligibility, under the same inadequate sound insulation at the contact between the conditions, iii) to perform listening tests (N=80) to penetrations and the building envelope and inadequate quantify the intelligibility of digits and subjectively assess pipe insulation, which does not prevent the transmission speech intelligibility using questionnaires (N=80), iv) to of external sound to the interior environment. compare the acquired speech intelligibility results acquired with the different approaches and to determine All these factors degrade the speech transmission. It has the suitability of the STI parameter as an indicator of been shown that students often have problems with speech intelligibility; and v) to develop recommendation hearing and understanding lectures; Čudina and Prezelj for speech intelligibility in lecture halls and classrooms [5] state that speech intelligibility in many classrooms and regarding the position and level of background noise lecture halls is less than 75%. This means that listeners levels. with normal hearing abilities understand less than 75% of randomly spoken words. The problem was exacerbated 2. METHODOLOGY during the Covid-19 pandemics, when lecturers had to wear face masks during lectures, and acrylic glass panels The STI measurements and the acquisition of the acoustic often obstructed the sound transmission path from the recordings were carried out in a lecture hall of a higher lecturer to the audience. Brown et al. [6] found that face education institution with a net volume of 280 m3 [Fig. 1], masks significantly affect speech transmission even at and a capacity of 40 listeners. In addition to STI, we moderate background noise levels. Thus, even at a Speech measured the reverberation time according to ISO 3382- to Noise Ratio (SNR) of -5 dB, which in simplified terms 2:2008 [8] and evaluated it in relation to the national means that the noise level is 5 dB higher than the speech technical guideline TSG-1-005:2012 [7]. level, speech intelligibility drops by 30%, both when using a cloth mask with a filter and when using transparent The recordings were made after the teaching activities on covers [6]. April 15 2022 in the 14:00-19:00 time range. The following measurement equipment was used: Despite the above research findings, the impact of • Reproduction of speech and of the STIPA signal building and equipment noise on the acoustic (hereafter referred to as artificial speaker): NTi Audio, environment in educational establishments is still poorly TalkBox, addressed in current design practice; the focus is only on • Spatial audio recording: 4th order ambisonics the requirements for maximum background noise levels, microphone: mh acoustics, Eigenmike®, not considering the importance of the position of noise • Class 1 audio analyser and level meter: NTI Audio, sources in the room. For example, for classrooms and XL2, lecture halls in Slovenia, TSG-1-005:2012: Noise Dovjak et al.: Determination of the position of equipment noise sources in educational institutions according to the subjectively evaluated speech intelligibility 63 AAAA – 2023 – IZOLA - Conference Proceedings • Dodecahedron loudspeaker for reproducing the to the nose level limits for more demanding intellectual HVAC noise and for reverberation time work defined by the Slovenian Guideline for Ventilation of measurements: NTi Audio, DS3 Dodecahedron (with Classrooms. NTI Audio, PA3 Power Amplifier), • Binaural reproduction: AKG, K702 (Reference Studio 2.1. Ambisonics recordings Headphones), • Computer audio interfaces: RME, FireFace UFX+ An audio sample of random digits was recorded in soundcard and PreSonus, HP60 headphone advance by four different speakers under studio preamplifier, conditions. This audio sample was reproduced by the • Diagnostic audiometer, Interacoustic, AD226. artificial speaker and recorded using an ambisonics microphone. The recordings were done under 26 conditions, differing in the level and position of the noise source. The recordings have been made with the microphone in two positions – in the first and last row (see Fig.1). In the next step, the digit recordings were processed to perform binaural listening tests. 2.2. STI measurements The Speech Transmission Index (STI) was measured under the same conditions and microphone locations as the ambisonics recordings. The STIPA signal defined by EN 60268-16 [11] was reproduced by the artificial speaker. For each condition, three STI measurements were taken, and finally, the mean value was used for the analysis. We obtained STI values for each investigated condition, i.e., Fig.1. The floor plan of the lecture hall used for the 26 in total. experiment (V=280 m3). The noise source position is indicated in red (A, B, C), the microphone positions are 2.3. Listening tests and subjective evaluation indicated in blue (first/last row), the artificial speaker was positioned at the location A. Subjective assessment of speech intelligibility was performed using a standardised questionnaire (5-12 May 2.1. The generated noise 2022) [12, 13], which consists of three parts. In the first part, seven general questions related to noise in lecture A dodecahedron loudspeaker was used to generate the halls were given; in the second part, two questions about artificially added HVAC noise. The height of the the effects that the respondents experience due to noise dodecahedron was fixed to 183 cm, while the ground were given; and in the third part, the participants listened positions were in points A, B and C (see Fig. 1). The noise to recorded audio samples of three random digits. For the was generated at four levels of 40 dB(A), 50 dB(A), 60 third part, the order of listening to digits recorded at dB(A) and 70 dB(A). The spectral characteristics of the different conditions has been randomized. The generated noise followed the spectrum of the Noise participants were asked to write down the digits and to Criteria Curves (NC), which are used to evaluate the evaluate the perceived intelligibility corresponding to the background noise of unoccupied buildings and rooms [9]. digits on the 0-10 scale. Since the ANSI/ASA S12.2-2008 standard for schools suggests an NC value between 25 and 35, the NC-30 curve We recruited 80 volunteers to carry out the questionnaire was used to generate the spectrum of the ventilation after which we also evaluated their hearing threshold. noise [10]. The lowest HVAC noise level also corresponds From the obtained audiograms, we identified a hearing Dovjak et al.: Determination of the position of equipment noise sources in educational institutions according to the subjectively evaluated speech intelligibility 64 AAAA – 2023 – IZOLA - Conference Proceedings impairment of one of the test subjects whose responses oriented more toward the recording position in the last were excluded from the analysis. This research was row. This results in a higher proportion of higher approved by the Ethics Committee for Research Involving frequencies in the acquired STIPA signal and, Human Subjects, University of Ljubljana (2012-2021; consequently, a higher STI, although the distance to the Ljubljana, 18 February 2022). artificial speaker is larger for the last row recording. It is expected that in larger lecture halls the STI would behave 3. RESULTS AND DISCUSSION differently and show the trend of decrease with distance from the source, due to the SNR distance decrease. This The measured reverberation time was 1.5 s in the 1000 Hz claim is supported by the results of Čudina and Prezelj [5], 1/3 octave band. The value is almost twice exceeding the showing that both the A-weighted speech level [dB(A)] recommended values for the room to be used for and the speech intelligibility index value decrease as one lecturing and it has been estimated that reverberation moves towards the last row. time would be too long even when the room is fully occupied. In fact, the optimal reverberation time ( T opt) for the room would be 0.61 s with the upper tolerance range at 0.74 s [7]. 3.1. Speech Transmission Index (STI) The results of the STI measurements shown (see Fig. 2) that the highest speech intelligibility is achieved in the cases with no additional noise and in the cases with additional noise at a level of 40 dB. Under these conditions, the STI values are around 0.5, which, according to BS EN 60268-16 [11], is considered as sufficient for rooms such as lecture halls (classes H and G). At the 50 dB noise level, the STI values range from 0.40 to 0.44 corresponding to the poor intelligibility (class I), while at the 60 dB noise level the STI values range from 0.24 to 0.28 corresponding to the very poor intelligibility (class U), which is also the achieved class for the 70 dB noise case, where the STI values range from 0.11 to 0.13. The 40 dB level coincides with the recommendation of the current Guideline for the Ventilation of Classrooms (IZS MSS-01/2021), which sets a limit of 40 dB(A) as the maximum noise level from air-conditioning systems for Fig.2. Measured STI for different positions and levels of the most demanding intellectual work [14]. The same HVAC noise and for the listening position in the first and limitation is also in the national Technical Guideline TSG- last row. 1-005 [7]. 3.2. Survey results The effect of the noise source position on the STI is relatively small, but there is a clear trend towards position Fig. 3 shows the distribution of responses on the five-level C being the most unfavourable. Similarly, we generally Likert scale acquired with the questionary. Part 1 included observe higher STI values for the position of the 7 questions: microphone in the last row. This is effect is most likely due • How likely is the specific noise experienced to to the directional properties of the speaker, which is influence the hearing capability of lectures? (a-g) Dovjak et al.: Determination of the position of equipment noise sources in educational institutions according to the subjectively evaluated speech intelligibility 65 AAAA – 2023 – IZOLA - Conference Proceedings Part 2 included 2 questions: • How often do you change the seat in the lecture hall due to noise? (h) and • How often noise influences your concentration? (i). The participants confirm that it is likely (mean 3.56) for their concentration to be influenced by noise sources (question i). Nevertheless, they rarely take action, i.e., change seat due to noise in the room (question h, mean 1.80). Among the listed noise sources none of them significantly stands out. Noise of the devices (option c) and building equipment (option d) are the least reported with the mean of 1.62 and 1.67, respectively. The remaining options are relatively with the mean values around 2.5. Fig.4. Total number of errors performed when listening 3.3. Number of errors to recorded digits at different microphone position, noise source position and generated noise levels. The total (cumulative) number of errors of the digit intelligibility is shown in in Fig. 4. As expected, the number Comparing the number of errors at the noise levels of 60 of errors increases with the noise level. At noise levels of and 70 dB in relation to the microphone position, we can 50 dB or less an insignificant number of errors has been see that a lower number of errors occurred in the last row. made by listeners, while at the high noise levels (70 dB) in In this row, is the number of errors also highest for the C most conditions all the 80 digits were unintelligible. In the noise source position. This indicates that the proximity to investigate noise levels range, the intelligibility was very the noise source at high levels strongly affects the discreetly transitioning from the full intelligibility to the intelligibility. full unintelligibility. To observe a smoother transition, it would be interesting to additionally investigate the The fact that more errors are present in the first row subjective intelligibility at intermediate noise levels. agrees with the trend observed for STI, which is favouring the recording/listening position closer to the axis of the speaker, i.e., in the last row. Nevertheless, no plausible explanation can be provided for the C position of the noise source at the 60 dB performing the worst, since the distances between the listener and the A, B, and C points is approximately equivalent. 3.4. Subjective rating – speech intelligibility assessment The acquired subjective ratings of speech intelligibility are presented in Fig. 5. Similarly to the STI and digits errors, the subjective speech intelligibility scores decrease most markedly as the level of additional noise increases. Alike, is also the overall intelligibility in the last row higher. Fig.3. Number of responses on part 1 and 2 of the questionary on the five-level Likert scale. The noise source position also influences the subjective speech intelligibility ratings. For the first row, the difference between A, B, C positions is relatively small, Dovjak et al.: Determination of the position of equipment noise sources in educational institutions according to the subjectively evaluated speech intelligibility 66 AAAA – 2023 – IZOLA - Conference Proceedings with the position C generally producing slightly lower On the other hand, at the higher noise levels of 60 and 70 score. More interesting results can be observed for the dB the position of loudspeaker C generates the lowest last row listening. In fact, in the cases with low additional perceived intelligibility. This means that at higher noise noise levels of 40 and 50 dB, the loudspeaker position has levels, the effect of the distance from the noise source and an effect depending on the direction of the speech and the related SPL becomes more relevant than the direction noise sources relatively to the listener; the more the two of sound sources. In this range, the results agree with directions differ, the higher the subjective intelligibility measured STI values, in which case loudspeaker position scores. The fact that such dependency is not observed in C is also the worst performing. the first row is most probably related to the listening position being further from the acoustic axis of the From the last row results it can be summarized that two speaker. This reduces the high frequency components of different regimes can be observed: a low noise regime, in speech and lead to a lower capability to discriminate which the direction of the noise source is more important, between sound sources. and a high noise regime, where the distance to the noise source is more important to achieve high intelligibility. For this reason, the best configuration for the last row We would expect that the observed regimes would be listening was C, in which case the noise source was most more pronounced if the investigated lecture hall would differing in direction compared to the speaker. This have a shorter acoustic response and as such facilitate the behaviour is not in agreement with the STI results which localisation and discrimination of sound sources. showed A noise source position as the best performing. It follows, that the different direction of the speaker and 4. CONCLUSION noise source cannot not be detected by STI which is in this sense a limited parameter. The intelligibility in a lecture hall has been investigated for HVAC noise sources at different locations and SPLs. The complex research approach included advanced sound recording and reproduction techniques, listening tests, and questionnaires with the goal of identifying factors that are relevant for speech intelligibility. On this basis design recommendations for introducing noise sources into the built environment could be developed. The main conclusions of the research are: 1. All intelligibility rating are most affected by the additional SPL level of the noise source. 2. STI indicators are also affected by the position of the microphone, showing that in the investigated room the position of the listener on the acoustic axis of the speaker is slightly favourable, which, in our case, was the case in the last row. 4. Considering the subjective assessment of intelligibility, the noise source furthest to the listener produces the best listening conditions at higher noise levels. In contrast, at low noise levels, moving the noise source away from the speaker is favourable. Fig. 5. Box plot of the subjectively rated speech 5. Overall, on the basis of the obtains results it is possible intelligibility on the 0-10 scale for different noise levels to formulate the recommendation that at lower noise and noise source position, when listening in the first levels positions away from the speaker should be chosen (top) and last (bottom) row. Dovjak et al.: Determination of the position of equipment noise sources in educational institutions according to the subjectively evaluated speech intelligibility 67 AAAA – 2023 – IZOLA - Conference Proceedings while at higher noise levels noise positions away from the Environment in Schools, Language, Speech and listener should be used. Hearing Services in Schools, 31, pp. 376 – 384, 2000. https://doi.org/10.1044/0161-1461.3104.376. [4.] ASHA. Classroom acoustics, American Speech- Generally, it can be concluded that the presented Language-Hearing Association, 2022. Available at: approach has produced valuable results. As part of future https://www.asha.org/public/hearing/classroom- research, the digit test should be performed at more noise acoustics/ (Accessed 8 July 2023). levels, to avoid the high quantisation of the obtained [5.] Čudina, M. and Prezelj, J. Intelligibility of speech in results. Furthermore, the research could be extended to classrooms and lecture halls, AR. Arhitektura, include a broader variety of noise sources that is present raziskave, 7(1), pp. 39 – 46, 2007. [6.] Brown, V. A., Van Engen, K. J. and Peelle, J. E. Face in the built environment. In addition, it would be mask type affects audiovisual speech intelligibility interesting to investigate, how different room acoustic and subjective listening effort in young and older conditions influence the obtained results. adults. Cognitive Research: Principles and Implications, 6: 492021. Acknowledgement https://doi.org/10.1186/s41235-021-00314-0. The authors acknowledge the financial support from the [7.] TSG-1-005:2012. Noise protection in buildings, 2022. Slovenian Research Agency (research core funding No. P2- Available at: https://www.gov.si/assets/ministrstva/MOP/Dokum 0158, Structural engineering and building physics, enti/Graditev/tsg_005_zascita_pred_hrupom.pdf research core funding No. Z1-4388, Toward better (Accessed 8 July 2023). understanding the diffuse sound field, and research core [8.] EN ISO 3382-2:2008. Acoustics – Measurement of funding No. J4-3087, Engineered wood composites with room acoustic parameters – Part 2: Reverberation enhanced impact sound insulation performance to time in ordinary rooms, 2008. improve human well being) and the European [9.] Noise curves, 2022. Available at: Commission for funding the InnoRenew project (grant https://web.archive.org/web/20180917215319/http s:/www.nti-audio.com/en/support/know-how/what- agreement #739574) under the Horizon2020 Widespread- are-noise-curves (Accessed 8 July 2023). Teaming program and the Republic of Slovenia [10.] ANSI/ASA S12.2-2008. Criteria for Evaluating (investment funding from the Republic of Slovenia and the Room Noise, 2008. European Union from the European Regional [11.] BS EN 60268-16:2011. Sound system equipment – Development Fund). Part16: Objective rating of speech intelligibility by speech transmission index, 2011. [12.] Ricciardi, P. and Buratii, C. Environmental quality 5. REFERENCES of university classrooms: Subjective and objective evaluation of the thermal, acoustic, and lighting [1.] Madbouly, A. I., Noaman, A. Y., Ragab, A. H. M, comfort conditions, Buildings and Enviroment, 127: Khedra, A. M. and Fayoumi, A. G. Assessment model 23-36, 2018. of classroom acoustics criteria for enhancing speech https://doi.org/10.1016/j.buildenv.2017.10.030. intelligibility and learning quality, Applied Acoustics, [13.] Jansen, S., Luts, H., Wagener, K. C., Frachet, B. and 114, pp. 147 – 158, 2016. Wouters, J. The French digit triplet test: Ahearing https://doi.org/10.1016/j.apacoust.2016.07.018. screening tool for speech intelligibility in noise. [2.] Serpilli, F., Di Loreto, S., Lori, V. and Di Perna, C. The International Journal of Audiology 49: 378-387. 2009. impact of mechanical ventilation systems on https://doi.org/10.3109/14992020903431272. acoustic quality in school environments in 52nd [14.] IZS MSS-01/2021. Guidelines for classroom AiCARR International Conference "HVAC and Health, ventilation, recommendations for the Comfort, Environment - Equipments and Design for implementation of effective mechanical ventilation, IEQ and Sustainability" E35 Web Conferences, 2021. Volume 343, 2022. https://doi.org/10.1051/e3sconf/202234305002. [3.] Sieben, G. W., Gold, M. A, Sieben, G. W. and Ermann, M. G. Ten Ways to Provide a High-Quality Acoustical Dovjak et al.: Determination of the position of equipment noise sources in educational institutions according to the subjectively evaluated speech intelligibility 68 Contributed papers Noise and vibrations 1. Integration of Psychoacoustic Perception for Enhanced Design of Axial Fans Nejc Cerkovnik (University of Ljubljana, Faculty of Mechanical Engineering) 2. Noise generating mechanisms analysis and its optimization on electronical commutated wet-dry vacuum cleaner suction unit Andrej Biček (Nela d.o.o) 3. Modeling and assessment wind turbine noise at different meteorological conditions Antonio Petošić (University of Zagreb, Faculty of electrical Engineering and Computing) 4. Evaluation of model based noise protection study based on in-situ vibro-acoustic railway track analy Krešimir Burnać (Faculty of Civil Engineering, University of Zagreb) 5. Steel reilway bridge noise, lack of reduction effect on airborne noise due to vibration dampers possibly acting a s noise sources. Rok Rudolf (ZAG) 6. Psychoaoustics of Pseudosound in Turbulent flow of Centrifugal Fan used in Household Appliance. Jurij Prezelj (University of Ljubljana, Faculty of Mechanical Engineering) 69 INTEGRATION OF PSYCHOACOUSTIC PERCEPTION FOR ENHANCED DESIGN OF AXIAL FANS Nejc Cerkovnik, Anže Železnik, Luka Čurović, Jure Murovec, Niko Tivadar*, Jurij Prezelj University of Ljubljana, Faculty of Mechanical Engineering *Corsair, Slovenia Abstract: Axial fans are widely used in cooling systems but often contribute significantly to environmental noise. Although optimizing the fan geometry usually leads to a reduction in sound power, related noise annoyance is usually not considered. Therefore, the geometry design focus was shifted towards optimizing the geometry for psychoacoustic perception. A psychoacoustic model is developed to correlate measured noise characteristics with subjective noise evaluations, enabling the objective psychoacoustic evaluation of fan prototypes. Psychoacoustic tests were conducted with 100 respondents evaluating recorded signals of noise from nine different axial cooling fans operating at different conditions. Psychoacoustic features were extracted from the recorded signals and correlated to survey results using linear regression analysis. This approach provided a robust model for future evaluation of fan prototypes noise annoyance. The results show that high-frequency components have a major impact on the perception of fan noise. The use of aero-acoustic features on fan blades is suggested. This study highlights the importance of considering subjective noise evaluations in the design of axial fans. By integrating psychoacoustic perception into the optimization process, cooling systems can be engineered to minimize noise annoyance. The developed psychoacoustic model serves as a valuable tool for evaluating axial cooling fan noise of various prototypes, providing insights for future fan design improvements in terms of both acoustic performance and cooling efficiency. Keywords: axial fan, acoustics, psychoacoustics, noise 1. INTRODUCTION comfort, productivity and even health. The consequences of noise annoyance go beyond mere acoustic discomfort In today's rapidly evolving world, which includes areas and can significantly affect various aspects of an such as artificial intelligence, cloud computing, and individual's well-being. advanced data processing, there is a growing global Noise pollution affects people's comfort indoors. Quiet demand for efficient computer operations. Fast data environments are sought out for relaxation, processing causes the temperature of the processor to concentration, and overall well-being. Unwanted noise, rise. Effective management of the accumulated heat is especially from continuous sources such as cooling fans, important to ensure processor reliability and can disrupt the desired atmosphere and hinder the performance. Within this thermal framework, the role of creation of comfortable environments that invite work, low-pressure axial fans as cooling components takes a rest, and social interactions. prominent position. While the design of traditional fans Noise can also directly impact productivity and has primarily focused on efficiency, the issue of noise performance, especially in areas where concentration and reduction has become significantly more important, cognitive tasks are critical. In workplaces where especially for fans used indoors where the user is directly professionals are engaged in critical thinking and problem exposed to the noise. solving, disruptive noise can impede cognitive processes, reducing efficiency and task performance. Noise pollution associated with computer cooling systems Prolonged exposure to noise can be detrimental to health. and axial fans can have a variety of effects that affect Chronic noise exposure, even at levels that do not cause Cerkovnik et al.: Integration of Psychoacoustic Perception for Enhanced Design of Axial Fans 70 AAAA – 2023 – IZOLA - Conference Proceedings hearing damage, has been associated with sleep models derived from recorded microphone signals are disturbance [1] the release of stress hormones [2], and constantly updated with new variables and features. consequently an increased risk of cardiovascular disease Standardised basic psychoacoustic parameters tonality T, [3]. In the context of indoor spaces with cooling systems, percentile loudness N5, roughness R, fluctuation strength individuals, particularly those who spend significant time F, and sharpness S exist for the evaluation of various in such spaces, may experience increased stress levels and technical sounds and are simply measured and calculated impaired sleep quality due to persistent noise pollution. [4]. While conventional approaches to fan design often focus Zwicker proposed a model to estimate psychoacoustic on improving efficiency and reducing sound power levels, annoyance from technical noise, such as household the psychoacoustic approach considers human perception appliances, but the model ignores the influence of tonal of noise. By integrating psychoacoustic principles into fan noise and therefore underestimates it [4]. Tonality is geometry optimization, noise annoyance can be mentioned as one of the main sources of annoying noise minimised, resulting in a more comfortable acoustic and usually occurs when the turbomachine operates environment. under optimal operating conditions. Guoqing [13] (Eq. 1- The main objective of our research is to establish a robust 4) extended the Fastl model [14] to include the tonality correlation between measured noise characteristics and compensation term wT and thus has been used in the subjective noise evaluations through the development of evaluation of axial fan noise, which can be highly tonal due an objective psychoacoustic evaluation model that could to the rotating nature of the machine. be used to evaluate optimised prototype geometries of low-pressure axial fans. Prior to this, listening tests must 𝑃𝐴 2 + 𝑤2 + 𝑤2) be conducted with a representative number of 𝐺𝑜𝑢𝑞𝑖𝑛𝑔 = 𝑁5 (1 + √𝑤𝑠 𝐹𝑅 𝑇 (1) participants to provide us with subjective ratings of (𝑆 − 1.75) 0.25 log(𝑁 𝑤 5 + 10) 𝑆 > 1.75 existing axial fan noise. 𝑠 = { (2) 0 𝑆 < 1.75 The basic ideas of psychoacoustics metrics and the 2.18 𝑤 (0.4𝐹 + 0.6𝑅) (3) evaluation of the quality of a product based on its sound 𝐹𝑅 = 𝑁0.4 5 were developed by Fastl and Zwicker [4] in the 1960s and 𝛽 𝑤 𝑇 𝛼 = 0.52 𝛽 = 6.41 (4) have become more widespread in the last two decades. 𝑇 = 𝑁𝛼5 Today, psychoacoustic metrics are used in the automotive industry [5,6], machinery fault diagnosis [7], in the Lipar et al. [15] have used experiments and jury-based household appliance industry [8-10], and in listening tests to develop an objective model for environmental noise monitoring [11]. psychoacoustics for use with vacuum cleaners and suction Listening tests based on the subjective evaluation of units. They have found that higher roughness levels are sound by the listener have always been used to assess preferable to achieve less disturbing sound and therefore sound quality. In the subjective evaluation of sound this effect was included in the model (Eq.5). quality, the use of semantic differentials is well known and has proven to be helpful. 100 100 1 However, because it depends heavily on the subjective 𝑃𝐴𝐿𝑖𝑝𝑎𝑟 = 0.96 ( + + + 𝐹 + 𝑅) (5) 𝑁 𝑆 𝑇 responses of the listener, the human response to noise is inconsistent and unreliable [12]. As a result, numerous Since different authors have presented models for tests are required, leading to time-consuming and different technical sounds, the proposed metrics differ expensive methods for sound quality assessment. from each other. To obtain an objective psychoacoustic Psychoacoustic parameters that can be measured and metric for low-pressure axial fans, subjective responses in standardized produce consistent results by streamlining the form of listening tests must also be included. processes and reducing costs. For any type of product, Therefore, the aim of this work is to validate the validation methods using listening tests and measured presented models and compare them with subjective psychoacoustic parameters are the best way to evaluations of cooling fans already performed and to demonstrate their value. To more accurately capture how develop appropriate metrics in case the existing ones are people perceive technical sounds, noise annoyance not sufficient Cerkovnik et al.: Integration of Psychoacoustic Perception for Enhanced Design of Axial Fans 71 AAAA – 2023 – IZOLA - Conference Proceedings 2. METHODOLOGY Subjective attained Calculated metrics from semantic differentials recordings Acoustic measurements were performed on 9 different Sound pressure level Lp Week – Powerful types of axial computer cooling fans. They were installed [dB] in pairs directly on the water cooling radiator in the Malfunctioning – well Loudness N computer case. The fans drew air from inside the case and 5 [sone] functioning blew it out through the water cooling radiator. To avoid Annoying – pleasant Shapness S [acum] interference from ambient noise, measurements were Cheap – expensive Roughness R [asper] performed in an anechoic chamber. Brüel & Kjaer and two Fluctuation strength F Rode NT5 microphones were used in combination with a Rough noise – soft noise [vacil] Brüel & Kjaer and Motu stage B-1 converter to record Sharp noise – damped sound signals with such low sound pressure levels. All of Tonality T [fone] noise the microphones were calibrated with a Brüel & Kjaer A weighted sound calibrator. Each pair of axial fans was recorded for 1 Tonal noise – broadband pressure level LpA in minute at various revolutions per minute (RPM) from a noise octave bands [dBA] distance of 1 meter above the computer case. The Fluctuating noise – experimental setup is shown in Figure 1. According to ISO constant noise 5801 [16], the airflow of each fan generated in the setup with the water cooling radiator mounted was measured. Table 1: Measured variables Listening test were conducted with 100 participants in the age group between 20 and 50 years. Stereo recordings from Rode microphones were replicated with headphones, and a special program was designed to guide participants through calibration, introduction with a series of examples of tonal-broadband, rough-soft, fluctuating- constant, and sharp-damped noise to prepare them for the test, and the final listening test, which consisted of 22 random order reproductions of axial fan recordings at two different rotational speeds. While listening to the recordings, subjects had to rate the sound they heard using semantic differentials on a scale of -5 to 5. They had to rate the quality of the sound subjectively in terms of the 4 differentials shown in the first column of Table 1. The second column in Table 1 contains 4 differentials that allowed participants to rate the subjective psychoacoustic Figure 1: Measurement setup metrics. After completion of the test, each individual's results were stored in a database. The semantic differentials used in the subjective listening tests and the 3. RESULTS AND DISCOUSSION objectively calculated metrics are summarized in Table 1. For the calculated metrics, the presented models (Eqs. 1 and 5) were used and compared with the subjective The aim of the linear regression between two selected semantic differentials was to find the correlations evaluation of the cooling fans using linear regression to between them and thus to emphasize the importance of determine the possible correlation between the two variables. The regression coefficient R2 was calculated to the selected metric in order to take it into account in the design of new geometries of axial fans. The results in indicate the strength of the correlation. Figure 2 show us otherwise. The subjectively evaluated noise annoyance was compared with the calculated sound pressure level (SPL). We see that the correlation is low – Cerkovnik et al.: Integration of Psychoacoustic Perception for Enhanced Design of Axial Fans 72 AAAA – 2023 – IZOLA - Conference Proceedings R2 value of 0.24. This result clearly shows that the noise of quiet cooling fans is not necessarily pleasant and emphasizes the importance of including psychoacoustic perception in the design process of cooling fans. Figure 2: Correlation between measured SPL and subjectively evaluated perception of noise as annoying – pleasant. Our results showed that the greatest effect on the sound pressure level and loudness of the axial fan is the amount of airflow delivered, an effect that is well known due to the intense vortex shading at the fan outlet [17]. Interestingly, the results showed that the subjective perception of a weak and strong noise is correlated with the amount of airflow conveyed. Figure 3A shows a correlation between airflow velocity, sound pressure level, and calculated loudness. When the linear regression function is positively sloped, it means that the loudness and SPL increase as the airflow velocity increases. Axial fans that produced higher airflow velocity were rated as more powerful by participants in the listening tests (Fig. 3C), and it was also found that the low-frequency component was related to both airflow velocity and the perception of the sound as weak or powerful (Fig. 3B). This implies that axial fans, which produce higher airflow for Figure 3: Correlation of delivered air flow of fans with the same system resistance, produce more intense vortex low frequencies components in noise signal end its shedding, increasing the overall sound pressure level, and subjective perception. are the source of the low-frequency components that contribute to the perception of the sound as more The only strong dependencies found between the powerful. Unfortunately, no correlation was found calculated standardised psychoacoustic metrics and the between powerful sound and the semantic differentials of subjective evaluation of sound quality were the annoying-pleasant and cheap-expensive. correlation of sharpness and the difference between annoying- pleasant noise. Sharp sound was found to be annoying, as shown in Figure 4. This finding needs to be Cerkovnik et al.: Integration of Psychoacoustic Perception for Enhanced Design of Axial Fans 73 AAAA – 2023 – IZOLA - Conference Proceedings directly addressed in future axial fan geometries to avoid annoying noise components. Figure 4: Correlation between sharpness and subjective assessment of noise quality in semantic differential of annoying noise – pleasant noise. Further investigation into what these components might be revealed that the higher frequencies have a major impact on noise quality. Fig. 5 shows the correlation between the high-frequency component at SPL and various subjective differentials. The participants rated the noise with emphasised high frequencies as cheap (Fig. 5 B), also high frequencies in the fan noise could represent Figure 5: Influence of high frequency components to a possible malfunction of the machine (Fig. 5 A) and subjective assessment of noise quality. therefore the overall noise was perceived as annoying (Fig. 5 C). Using psychoacoustic metrics from sound signals recorded with microphones, we calculated acoustic annoyance with Guaqing and Lipar models. Due to the large influence of the high-frequency components, we also developed a new objective psychoacoustic model. The acoustic annoyance was calculated using equation 6. 𝑃𝐴𝐻𝐹𝑅𝐸𝑄 = 0.05 ∗ 𝑇 − 0.05 ∗ 𝐻𝐹𝑅𝐸𝑄 (5) Where HFREQ is the sum of SPL at centre of the one third octave bands with frequencies 4000, 5000, 6300, 8000, 10000, 12500 and 16000 Hz and T is standardised tonality. The calculated values were plotted against the subjective evaluation of noise quality, more precisely against the semantic differential of annoying- pleasant noise. Linear regression showed us that Liparś and the newly proposed model provide results with strong agreement with subjective perception (Fig. 6B and 6C). The modified Cerkovnik et al.: Integration of Psychoacoustic Perception for Enhanced Design of Axial Fans 74 AAAA – 2023 – IZOLA - Conference Proceedings Guaqing model does not provide useful results, as shown Figure 6: Guaqing’s (A), Lipar’s (B) and newly proposed in Figure 6 A. (C) model for objective psychoacoustic annoyance compared to subjective noise assessment. 4. CONCLUSION The results of the conducted listening tests provide very interesting results. First of all, it was proved that the integration of psychoacoustic perception is important in the design of axial fans and that the single value of SPL is not enough to describe the sound as pleasant. It was found that axial fans producing higher airflow rates for the same system resistance of the water-cooled radiator had higher SPL levels measured and that the low frequency band (between 250 and 800 Hz) in the signal was more pronounced due to more intense vortex shedding. This is consistent with the subjective evaluation of the noise by the participants in the listening tests as being powerful. In addition, a large impact of high-frequency components (from 4 kHz to 16 kHz) was found on various subjective sound quality metrics. Sound signals containing the mentioned components were perceived as cheap and annoying. According to participants listening to the tests, such axial fans would be considered malfunctioning. In addition, sharp noises were considered annoying. All of the above issues must be carefully considered when designing axial fans. A simple solution to lower the overall level SPL and eliminate low frequencies would be to reduce the airflow generated, but with this change the axial fan loses its cooling capacity. Therefore, the focus must shift to eliminating higher frequencies from the noise. This could be accomplished by using aeroacoustics features such as serrated trailing edges [17], Gurney flaps [18], winglets [19], and swept blades [20]. Finally, the study has also shown that we can confidently use Lipar’s and, in particular, the newly proposed model for a rapid assessment of objective noise annoyance in new axial fan designs without having to perform subjective listening tests with a large number of participants. 5. REFERENCES [1.] Miedema H. M., Vos H. Associations between self- reported sleep disturbance and environmental noise based on reanalyses of pooled data from 24 studies, Behav. Sleep Med., 5(1), pp: 1-20, 2007, doi: 10.1207/s15402010bsm0501_1. Cerkovnik et al.: Integration of Psychoacoustic Perception for Enhanced Design of Axial Fans 75 AAAA – 2023 – IZOLA - Conference Proceedings [2.] Babisch W. Stress hormones in the research on based on psychoacoustics metrics, Applied cardiovascular effects of noise, Noise Health, 5(18), Acoustics, 157, 2020. pp: 1-11, 2003, PMID: 12631430. [11.] Kang J., Zhang M. Semantic differential analysis [3.] Sivakumaran K., Ritonja J. A., Waseem H., Al of the soundscape in urban open public spaces, Shenaibar L., Morgan E., Ahmadi S., Denning E., Build. Environ., 45(1), pp: 150 – 157, 2010. Michaud D.S., Morgan R. L. Impact of Noise [12.] Fastl H. The psychoacoustics of sound-quality Exposure on Risk of Developing Stress-Related evaluation. Acustica-Acta, Acustica 1997. Health Effects Related to the Cardiovascular [13.] Guoqing D. An improved psychoacoustic System: A Systematic Review and Meta-Analysis, annoyance model based on tonal noises, Hamburg, Noise & Health, 114, pp: 107-129, 2022. Germany, 2016 [4.] Fastl H, Zwicker E. Psychoacoustics – Facts and [14.] Fastl H. Hearing sensations and noise quality Models, 3rd ed. Heidelberg, Springer-Verlag, Berlin, evaluation, J. Acoust. Soc. Am., 87(S1), pp: 134-134, 2007. 1990. [5.] Arana M., San Martín R., Urroz J. C., Diéguez P. M., [15.] Lipar P., Prezelj J., Rejec J., Čudina M. Gandía L. M. Acoustic and psychoacoustic levels Psychoacoustic approach used for developing the from an internal combustion engine fueled by model of sound pleasantness of vacuum cleaners hydrogen vs. gasoline, Fuel, 317, 2022, and suction units based on objective and subjective https://doi.org/10.1016/j.fuel.2022.123505. analysis, 5th Congress of Alps-Adria Acoustics [6.] Leite R.P., Paul S, Gerges S. N. Y. A sound quality- Association, 12–14 September (2012), Petrčane, based investigation of the HVAC system noise of an Croatia. automobile model, Appl. Acoust., 70(4), 2009, [16.] ISO 5801: 1997. Industrial fans – performance https://doi.org/10.1016/j.apacoust.2008.06.010. testing using standardised airways Elsevier BV. [17.] Qingyi S., Haodong X., Jian C., Jiangtao [7.] Kane P.V., Andhare A.B. Critical evaluation and Z. Experimental and numerical study on a new noise comparison of psychoacoustics, acoustics and reduction design for a small axial fan, Applied vibration features for gear fault correlation and Acoustics, 211, 2023, classification, Measurement, 154, 2020, https://doi.org/10.1016/j.apacoust.2023.109535. https://doi.org/10.1016/j.measurement.2020.10749 [18.] Xuemin Y., Nan Z., Jiami H., Chunxi L., Zhanpu X. 5. Numerical investigation of the benefits of serrated [8.] Jeon Y.H., You J., Chang H.Y. Sound radiation and Gurney flaps on an axial flow fan, Energy, 252, sound quality characteristics of refrigerator noise in 2022, real living environments, Appl. Acoust., 68 (10), https://doi.org/10.1016/j.energy.2022.124072. 2007, [19.] Ye X., Zhang J., Li C. Effect of blade tip pattern https://doi.org/10.1016/j.apacoust.2006.06.005. on performance of a twin-stage variable-pitch axial Elsevier BV. fan, Energy, 126, pp: 535-563, 2017, [9.] Novaković T. et al. Psychoacoustic approach on https://doi.org/10.1016/j.energy.2017.03.057. centrifugal fan blade design, Fourth International [20.] Park M., Lee D., Lee H. Experimental and Acoustics and Audio Engineering Conference, 2017. computational investigation of the effect of blade [10.] Novaković T., Ogris M., Prezelj J. Validating sweep on acoustic characteristics of axial fan, impeller geometry optimization for sound quality Applied Acoustics, 189, 202, https://doi.org/10.1016/j.apacoust.2021.108613 Cerkovnik et al.: Integration of Psychoacoustic Perception for Enhanced Design of Axial Fans 76 NOISE GENERATING MECHANISMS ANALYSIS AND ITS OPTIMIZATION ON ELECTRONICAL COMMUTATED WET-DRY VACUUM CLEANER SUCTION UNIT Andrej Biček1, Janez Luznar 2, Igor Markič1 1 Nela d.o.o., Na plavžu 79, 4228 Železniki, Slovenia 2 Domel d.o.o., Otoki 21, 4228 Železniki, Slovenia 1 andrej.bicek@domel.com1 , j.luznar@domel.com2, igor.markic@domel.com1 Abstract: Vacuum cleaner’s market in EU has after cancelation of Energy labelling of vacuum cleaners nowadays face with the revised regulation which is expected to be started in autumn 2023. The Regulation will establish eco-design requirements for electric mains operated vacuum cleaners including hybrid vacuum cleaners. Domel as leading European producer of vacuum cleaner suction units today produce over 75% of those units in range of 600 to 900 watts of input electric power. Important share of production present electronical commutated motors and vacuum cleaner suction units for commercial and industrial applications. Demanding mounting conditions of those motors into devices and heavy operating duties requires introduction of new development methods and new technologies in production. Through this paper is presented systematic approach of detection of sound mechanisms and their optimization on electronical commutated wet-dry vacuum suction unit will be presented. Keywords: vacuum cleaner suction unit, electronical commutation, noise generation mechanism, sound power level, energy labelling, impeller geometry, rotor mass unbalance, vibrations, vacuum cleaner 1. INTRODUCTION The growing focus on promoting the use of energy- efficient home appliances in Europe has been driven the The worldwide vacuum cleaner market revenue in year market for advanced household vacuum cleaners. 2023 is planned on €40,89 billion, but if we divide on The Eco-design recommendations and Energy labelling individual markets €8,27 billion in the European market regulations for vacuum cleaners were started in 2014 with incomes are expected on and on China market electric power limitation to 1600W. At the beginning of €12,1billion [1]. 2017 the energy consumption of vacuum cleaners has been further limited to 900W, both regulation steps have been presented in paper by Biček et al. [2], [3] In November 2018 the Energy Labelling Regulation was annulled by General Court of the European Court of Justice (Case T-544/13 RENV). The regulation has been reviewed in 2019/2020 and is expected which is expected to be started in autumn 2023.[4] Vacuum cleaners (in following the instead name Vacuum cleaner abbreviation VC will be used) in general can divided into the following types: upright VC, handheld VC, canister VC, stick VC, robot VC, wet/dry VC, central VC, Fig. 1. Vacuum cleaner market revenue in period 2018- backpack VC. 2028 for different markets [1] Andrej Biček et al.: Noise generating mechanisms analysis and its optimization on electronical commutated wet-dry vacuum cleaner suction unit. 77 AAAA – 2023 – IZOLA - Conference Proceedings The topics that were reviewed are include: 2. AN REVIEW OF PAST WORK • Tolerance set in the verification procedure for Emitted noise of vacuum cleaner suction unit (VCU) is market surveillance purposes. consisting of aerodynamic and structural borne origins. • Whether full-size battery-operated vacuum Structural noise could be divided on mechanical, vibration cleaners should be included in the scope • and electromagnetic origins [6] whether it is feasible to use measurements methods based on a partly of loaded rather than Čudina and Prezelj [7] are in 2006 described some noise empty canister [4]. origins of dry suction unit. The authors found that trend of Following types of vacuum cleaners or similar device will vacuum cleaner development show speed increase from be out of scope of Energy labelling regulations: wet, wet 50000 min-1 to 70000 min-1 or even higher, meanwhile &dry, battery-operated, robot, industrial and central VC, lifetime of vacuum suction units is decreasing and level of floor polishers and outdoor VC. [4] emitted noise is opposite increasing. Further they Nevertheless, the buyers demand that product are describe origins of rotation and non-rotation noise and developed according to Eco-design recommendations: explain airflow and noise conditions in rotor depend of energy efficient, high efficiency of dust management and operating point. emitted low noise. Important addition is now also use of Čudina and Prezelj further in article [8] systematic discuss ecological recycling materials and carbon footprint effect of presence of blade and blades diffuser on analysis. All these restrictions and recommendations are aerodynamic performances and on noise level in closed important also for vacuum cleaner unit producers and acoustic field and its direction. They measured sound Domel is one of the biggest in Europe. pressure on grid on vacuum suction surface. They found that sound pressure level is oriented and therefore is needed to consider that to use same positions at noise measurements. Further they measured sound pressure level of VCU at different speeds and loads. Čudina and Prezelj in third article [9] performed analysis of structural excited noise on total noise level of VCU. They determined several positions on VCU surface and measured vibrations and sound pressure in the same. Through vibration velocity spectrum they determined noise origins regarding to frequency of rotation fR and blade passage frequency fRR. Additional they compared noise spectra of VCU with and without rotor of centrifugal blower (CB). They found that noise of electromotor alone is much lower compare of noise of centrifugal blower. At end authors concludes that dominant noise of VCU is Fig. 2. Types of vacuum cleaners which will be subjected aerodynamic origin, meanwhile electromagnetic origin to energy labelling, left – canister vacuum cleaner with bag, right – upright cyclone cleaner [5] depends on electromotor type, its geometry and load is main source of structure noise. Prezelj and Čudina in next article [10] identify part of aerodynamic and structural excited noise. They setup experiment on way that fixed VCU on device (see figure 4). Another VCU generated airflow through observed VCU, and they measured generated aerodynamic noise at blocked rotor. Structural noise they generated by mini shaker which excited housing of VCU. Result was transfer Fig. 3. Types of vacuum cleaners which are excluded function of aerodynamic and structural noises depend on from energy labelling: left - battery hand cleaner, middle - floor polisher, right - ash and grass suction frequency. unit [2] Andrej Biček et al.: Noise generating mechanisms analysis and its optimization on electronical commutated wet-dry vacuum cleaner suction unit. 78 AAAA – 2023 – IZOLA - Conference Proceedings Fig. 4. Aerodynamic and structural noise experiment [11] Ob structural noise could significantly influence rotor unbalance as rotational origin of VCU. Mass unbalance of rotor could be cause by design, material inhomogeneity, parts manufacture and assembly process [12]. Regarding to occurrence the rotor unbalance could be described as static, moment and dynamic unbalance. Petrič et al. [13] describe problematic of unbalance rotor of alternator. They found that material cut out gaps on surface rotor due of balancing process collapse symmetry of geometry shape and consequently also magnetic Fig. 5. Sound generating mechanisms of vacuum cleaner symmetry of rotor because of local increased air gap suction unit. [14] between rotor and stator pole or magnet segments. By 2D finite element method (FEM) authors have been 3.1. Magnetic noise of electromotor demonstrated that reduced density of magnetic field effect on reduced magnetic force. Experimental have Magnetic noise is occurred due of periodic excitation of authors confirmed that number and depth of gaps effect electromotor structure with magnetic forces and the on vibrations. Measurements on motor bracket have phenomenon of magnetostriction. In the latter due shown that vibration levels before and after balancing variable magnetic field, the elastic deformation in remained at equal level, sometimes has been even material is occurred and this resulting in magnetostriction increased. noise. This is typical for large lamination structures like the transformers, meanwhile for the medium or small 3. SOUND GENERATING MECHANISM OF VACUUM electromotors this effect is usually neglected. CLEANER SUCTION UNIT Sound generating process is complete system consist of 3 basic components like as sound origin, the path of propagation and noise detection (by listener or measurements). Starting point of noise generation are caused with several excitation forces through complex connections which are for vacuum cleaner suction unit in general listed in Fig. 5. Elementary noise origins could divide on aerodynamics, mechanical and magnet sources, which effect through parts main function or parts geometry deviation generation forces on electromotor elementary lamination of structure (housing) of suction cleaner motor. Ratio of noise sources could be different regarding to type of electromotors like brush commutated electromotor or electronically commutated motor. Fig. 6. Mechanism of noise and vibration generation in electromotors[15] Andrej Biček et al.: Noise generating mechanisms analysis and its optimization on electronical commutated wet-dry vacuum cleaner suction unit. 79 AAAA – 2023 – IZOLA - Conference Proceedings Electromagnetic forces can be shared on useful tangential 3.2. Structural dynamics forces that cause torque and radial forces that act on stator of electromotor and cause radial deformation and noise as shown Fig. 6 [15] Last period has been on market increased ratio of electromotors controlled with electronical commutation, which is process of sequence control relative on rotor position. Electronical commutation in this case fulfils traditional task of commutator and brushes with main difference that for commutation or switching of inductive windings control electronics is used. Complete scheme of electronically commutated electromotor from the input electric power supply to the mechanical output as torque is shown of Fig. 7. Fig. 9. Experimental rotor eigenfrequencies identification Excitation forces effect on structure dynamics of electromotor parts and they should be tested to avoid later issues at electromotor operation. In development process are used mechanical simulations with Ansys Fig. 7. Electronically commutated motor in 3 segments as input, electronics, motor [16] software and results are subsequently confirmed with experimental modal test as it shown for rotor on Fig. 9 Fig. 7 show electronics with 6 transistors care that electromotor is controlled with pulse – wide modulation 3.3. Aerodynamic noise (PWM). The PWM excitation harmonics cause electromagnetic forces, which interact with the structure. The PWM voltage excitation harmonics depend on the carrier type, the carrier frequency fc, the fundamental frequency f1 and the modulation index ma ,generated frequency harmonics are shown on Fig. 8 .[17] The variable motor load results in different proportions of the switching harmonics in the in the PWM voltage excitation and therefore the structure-borne noise minimization is complex task Fig. 10. Simulation tool for mechanical and aerodynamic optimization of parts of vacuum cleaner suction unit Aerodynamic noise is occurred due high speed and nonstationary flow of air through aerodynamic part of vacuum cleaner suction unit and interaction of airflow with obstacles. Geometry of impeller (rotor) and diffuser Fig. 8. Frequency contents of the voltage excitation (stator) ensure performance parameters on other side harmonics for a sine-triangle PWM [17] important effect also on acoustics behave. Therefore, beside experiences also simulation software like Ansys Fluent are used, simulation steps are presented on Fig. 10 Andrej Biček et al.: Noise generating mechanisms analysis and its optimization on electronical commutated wet-dry vacuum cleaner suction unit. 80 AAAA – 2023 – IZOLA - Conference Proceedings 4. OBJECTIVES AND EXPERIMENT rotors with laser material removal as is presented on Fig. 12. 4.1. Electronically commutated vacuum suction unit - (EC VCU) Vacuum suction unit VCU 467 and its optimization has been presented in some previous papers. [18] Eco design goals to developed energy efficient and less material weight products Domel used to developed new EC VCU 759 as it presented in Fig. 11. Fig. 12. Rotor balancing process with nonconventional technology. Unbalance values of group of rotors and impellers prepared for test are presented on Fig. 13 and Fig. 14. Fig. 11. Vacuum cleaner suction units (VCU) comparison: VCU467 and EC VCU759 4.2. Experiment design At VCU optimization we designed experiment to analyse effect of input unbalance of rotors and impellers on vibrations on VCU. Test has been started with 2 groups of rotors and 2 groups of impellers (non-dimensional values of unbalance is presented in table 1) which has been separate prepared on balancing machine. Fig. 13. Unbalance values for rotors at balancing planes Table 1. Unbalance values of rotors and impellers for test U1 and U2 Unbalance value [/] Group of EC Rotor unbalance Impeller unbalance VCU 759 1 0,45-0,75 >0,6 2 0,45-0,75 <0,45 3 < 0,3 >0,6 4 <0,3 <0,45 4.2. Balancing process High requirements regarding low vibration levels of EC VCU require that rotors have low unbalance values. Domel developed new technology approach of precise balancing Fig. 14. Static unbalance values of impellers Andrej Biček et al.: Noise generating mechanisms analysis and its optimization on electronical commutated wet-dry vacuum cleaner suction unit. 81 AAAA – 2023 – IZOLA - Conference Proceedings 4.3. Vibration measurements on assembly line preload, axial magnetic forces, and impeller axial deviation. 6. CONCLUSION Vibration RMS velocity scaled values were presented for 4 groups of EC VCU with 2 groups of balanced rotors and impeller. At development of new EC VCU generation new materials has been used therefore also new modelling approached has been needed. To achieve sharp requirements 2 new technologies in process like laser welding and balancing have been implemented. Issues will be used at new projects. Fig. 15. Diagnostic device on assembly line for vibration 7. REFERENCES control EC VCU has been assembled on full automatized assembly [1] Vacuum Cleaners - Europe | Statista Market line and inspected on 3 stage diagnostics control devices Forecast, on end of assembly line. Vibrations on this device are https://www.statista.com/outlook/cmo/househo measured with laser vibrometer Polytec IVS 500. ld-appliances/small-appliances/vacuum- For each EC VCU vibrations has been measured on 3 cleaners/europe?currency=eur, (n.d.). positions on one section plane like axial on cover, radial https://www.statista.com/outlook/cmo/househo on cover and radial on middle of stator lamination. ld-appliances/small-appliances/vacuum- cleaners/europe?currency=eur (accessed 5. RESULTS September 18, 2023). [2] E. Commission, Guidelines accompanying : Commission Delegated Regulation ( EU ) No 665 / 2013 of 3 May 2013 supplementing Directive 2010 / 30 / EU with regard to energy labelling of vacuum cleaners and Commission Regulation ( EU ) No 666 / 2013 of 8 July 2013 implementi, (2014). [3] Andrej. Bicek, Igor. Markic, Janez. Rihtarsic, Jozica. Rejec, Mirko. Cudina, Vacuum cleaner suction unit performance and noise characterization an overview, 6th Congress of the Alps Adria Acoustics Association, October 16th and 17th 2014, Graz. (2014). Fig. 16. Vibration velocity RMS values for vibration in [4] European Commission: Ecodesign and Energy axial, radial position on cover and radial position on Labelling Working Plan 2022-2024, housing (stator lamination) commission_guidelines_ecodesign_requirements Vibration levels on Fig. 16 show that there is present _for_vacuum_cleaners, (n.d.). correlation between unbalance and vibration values. commission_guidelines_ecodesign_requirements According to input unbalance values on Table 1 we can see _for_vacuum_cleaners (accessed September 18, that radial vibration slightly in average slightly drop down. 2023). At axial vibration there is no correlation which is logical [5] Electrolux Company website, due different vibration mechanism, like axial spring https://www.electrolux.si/vacuums-home- comfort/vacuum-cleaners/, (n.d.). Andrej Biček et al.: Noise generating mechanisms analysis and its optimization on electronical commutated wet-dry vacuum cleaner suction unit. 82 AAAA – 2023 – IZOLA - Conference Proceedings [6] M. Čudina, Tehnična Akustika I, 2011. rotorja glede na vibracije enosmernih [7] M. Čudina, J. Prezelj, Noise generation by vacuum elektromotorjev, Strojniski Vestnik/Journal of cleaner suction units. Part I. Noise generating Mechanical Engineering. 49 (2003). mechanisms - An overview, Applied Acoustics. 68 [14] M. Furlan, Karakterizacija magnetnega hrupa (2007). enosmernega elektromotorja = [Magnetic noise https://doi.org/10.1016/j.apacoust.2006.10.003. characterization in the DC electric motor] : [8] M. Čudina, J. Prezelj, Noise generation by vacuum doktorsko delo, 2003. cleaner suction units. Part II. Effect of vaned [15] J. Luznar, Vibroakustična karakterizacija diffuser on noise characteristics, Applied elektronsko komutiranih motorjev: doktorsko Acoustics. 68 (2007). delo, [J. Luznar], 2019. https://repozitorij.uni- https://doi.org/10.1016/j.apacoust.2006.10.002. lj.si/IzpisGradiva.php?id=113188. [9] M. Čudina, J. Prezelj, Noise generation by vacuum [16] Fluke Inc., “How to measure output voltage from cleaner suction units. Part III. Contribution of a VFD to a motor” , [Online]. Available: Http://En- structure-borne noise to total sound pressure Us.Fluke.Com/Community/Fluke-News- level, Applied Acoustics. 68 (2007). plus/Motors-Drives-Pumpscompressors/How-to- https://doi.org/10.1016/j.apacoust.2006.10.001. Measure-Output-Voltage-from-a-Vfd-to-a- [10] Jurij. Prezelj, Mirko. Čudina, Quantification of Motor.Html. (2012). aerodynamically induced noise and vibration- [17] J. Luznar, J. Slavič, M. Boltežar, Experimental induced noise in a suction unit, Proc Inst Mech Eng research on structure-borne noise at pulse-width- C J Mech Eng Sci. 225 (2011) 617–624. modulation excitation, Applied Acoustics. 137 [11] J. Prezelj, M. Čudina, Quantification of (2018). aerodynamically induced noise and vibration- https://doi.org/10.1016/j.apacoust.2018.03.005. induced noise in a suction unit, Proc Inst Mech Eng [18] A. Biček, I. Markič, J. Rihtaršič, J. Rejec, M. Čudina, C J Mech Eng Sci. 225 (2011). Influence of cover outlet geometry on noise https://doi.org/10.1243/09544062JMES2187. spectra of wet vacuum cleaner suction units, in: [12] ISO 1925:, Mechanical vibration – Balancing – Hrvatsko akustičko društvo, 2012: p. NOI-03. Vocabulary,” Int. Organ. Stand., 2001. [13] M. Petrič, M. Furlan, M. Boltežar, Primerjava masnega ter magnetnega neuravnoteženja Andrej Biček et al.: Noise generating mechanisms analysis and its optimization on electronical commutated wet-dry vacuum cleaner suction unit. 83 PAPER TITLE: MODELLING AND ASSESSMENT WIND TURBINE NOISE AT DIFFERENT METEOROLOGICAL CONDITIONS Antonio Petošić, Domagoj Stošić, Mia Suhanek University of Zagreb, Faculty of Electrical Engineering and Computing, Department of Electroacoustics, Unska 3, Zagreb Abstract: In this paper, a model for predicting wind turbine noise under different meteorological conditions (i.e., wind speed magnitudes and directions at 10 metres height) is shown, and residual noise is measured at several different locations around the wind turbine site. The CONCAWE meteorological model method is used for estimating the influence of meteorological parameters with different meteorological classes on noise propagation from the wind turbine as a noise source with different sound powers at different wind speed magnitudes. The modelling results and uncertainties are discussed in terms of acquired meteorological and residual noise levels at considered micro locations around the site. The limit rules for those types of sound sources are analysed for the day/evening/night period regarding the distance from the houses and residual noise at different wind speed magnitudes. Regarding the noise legislation, the measures for reducing wind turbine noise and its influence on energy production are discussed. Keywords: wind turbine noise, wind speed and direction, sound power, limit rules, measures for reducing wind turbine noise 1. INTRODUCTION (>=500m) from the resident objects due to low sound attenuation of lower frequencies. Noise levels estimation from array of wind In this paper the measurement of residual noise turbines is a complex task due to their different sound at different wind speeds (due to aerodynamic noise power levels at different wind speeds and influence of around object) and wind directions during the meteorological conditions (favourable, neutral, representative period (together with assumed cloudiness unfavourable) on excess attenuation due to for the most favourable propagation meteorological class) meteorological conditions (Amet) [1,2]. In Croatia there is are done in duration of two months according to ISO 1996- a rigorous rule that the levels from new sound source(s) 2:2017 and ISO 1996-1:2016 [5,6]. The residual noise has are not allowed to increase the residual noise levels more been measured at critical locations in all wind directions than +1 dBA or the level of new installed sources should (bins for wind direction in 30o and bins for wind speeds in be 5 dBA less than proposed limit value [2,3]. 1 m/s intervals) around the planned wind turbine In the rural parts of the country (Croatia) the level installation locations. After the measurements of initial of residual noise is usually between 20-30 dBA during the state for residual noise from old wind turbine installed night depending on the wind speed at 10 metres and at nearby the sound pressure level LA,eq from new sources hub height. The sound power of wind turbines installed at have been calculated using appropriate modelling some height (100-200m) can be very high (>95 dBA) at software while changing the input parameters which lower wind speeds (3 m/s at 10m) and it is very difficult determine the propagation conditions to the resident to obey this rule for increase the residual noise for +1 dBA objects around the old and new wind noise farm. when the wind turbine is installed at a small distance The modelling results for LA,eq have been compared with residual noise levels ( LA,eq,res) at each direction (mid of Petošić at at. MODELLING AND ASSESSMENT WIND TURBINE NOISE AT DIFFERENT METEOROLOGICAL CONDITIONS 84 AAAA – 2023 – IZOLA - Conference Proceedings the bin) and wind speed (mid of the bin). The limit rule for increasing the residual noise by not more than +1 dBA has been tested according to eq. 1. 𝐿𝑟𝑒𝑠𝑖𝑑𝑢𝑎𝑙 𝐿𝑤𝑖𝑛𝑑𝑡𝑢𝑟𝑏𝑖𝑛𝑒 𝐿𝑡𝑜𝑡𝑎𝑙 = 10 ∙ 𝑙𝑜𝑔10 (10 10 + 10 10 ) (1) The sound power levels of wind turbines (testing different types, numbers, and locations) in one third octave bands are given for different wind speeds and Fig.1. Measuring and modelling situation with critical modes of working. The sound pressure levels are measurement positions and old and new sound sources calculated at different windspeeds directions (in bins 30 (8 wind turbines) degrees) and the wind speeds when it starts working up to limit wind speed. After analysing the residual noise levels at The modelling tool used in calculations was ISO 9613- different wind directions and speed at different critical 2:1997 [7] with CONCAWE meteorological model [1,2] measurement positions the sound pressure levels from This calculation procedure is included in the estimation of old wind turbine sources is calculated with known sound sound pressure levels from old (calibration procedure for powers and the model parameters (reflection coefficient critical directions) and new noise sources installed for and wind speed at hub height-roughness has been each wind speed, direction and meteorological classes calibrated). The residual noise level for the dominant (Pasquill) [1,2]. It is assumed in determination of propagation direction at 4 m/s speed (at 10 m height) has appropriate Pasquill stability classes that in that area zero been compared with modelling result for old wind cloud octas were present during measurements- the turbines at closest positions (MM1 and MM3) for the worst meteorological class conditions for propagation. appropriate wind speed at hub (calculated by assuming We have tried to use some other calculation models logarithmic wind speed profile and measured one) and (Harmonoise, CNOSSOS-EU [8]). However, only in average difference was in the range of ±0,5 dB with available programme support with ISO 9613-2:1997 ground reflection coefficient G=0.2. At higher speeds and calculation procedure the CONCAWE meteorological further locations, the dominant part is aerodynamic noise correction was available. According to the previous around objects, so the results are not used in calibration research the CONCAWE model is most suitable to assess of the model. Amet while estimating sound pressure levels. 2. THEORETICAL PROPAGATION MODELL 1.1. Measurement and modelling situation The ISO 9613-2:1997 calculation model [7] assumes the that the sound power, directivity, reflection Measurement and modelling situation with installed coefficient from ground (G=0.2), attenuation in air and wind turbine sources (old and new) is shown in Fig. 1. meteorological correction in frequency range of interest (usually octave) is known to find sound pressure levels at different positions. The calculation formula for sound pressure Lp is given by eq. 2, 𝐿𝑝 = 𝐿𝑤 − 20 ∙ 𝑙𝑜𝑔10(𝑟) − 11 (𝑑𝐵) + 𝐷 − ∑ 𝐴 𝑎𝑡𝑡𝑒𝑛𝑢𝑎𝑡𝑖𝑜𝑛 (𝑑𝐵) (2) where Lw is sound power of the source (usually in one third octave bands) r (m)-distance from the source D (dB)-directivity factor ( D=0 at hub height, point source) Petošić et al.: MODELLING AND ASSESSMENT WIND TURBINE NOISE AT DIFFERENT METEOROLOGICAL CONDITIONS 85 AAAA – 2023 – IZOLA - Conference Proceedings Attenuation factor depends on the many 3-4 26,9 28,9 33,4 24,2 30,6 30,6 - - - - 33,1 32,8 * * * complicate factors for each propagation frequency as 4-5 40,9 37,1 35,7 36,3 35,8 - - - - - 38,4 38,5 * * * given in eq. 3. 5-6 47,1 47,5 42,9 - 41,5 42,8 - - - - 43,1 43,7 * * * 6-7 51,7 44,8 44,7 45,6 47,5 - 46,2 - - - - - * * * * * A =A +A +A +A +A +A_ +.. (3) 7-8 51,6 44,4 48,6 51,8 53,7 - - - - - - - attenuation atm ground obstacle vegetation meteorology turbulence * * * * * 8-9 61,1 58,2 56,7 - - - - - - - - * * * - The most critical factor in wind turbine noise is 9-10 64,8 62,1 58,7 - - - - - - - - - * * * meteorology Amet which depends on the wind speed and 10-11 62,8 63,2 - - - - - - - - - - * * direction (wind speed component in the source-receiver 62,4 11-12 - - - - - - - - - - - * direction) and on the solar radiation. The detailed *Only few samples averaged connection between Pasquill stability class and Noise levels/ dB-nighr meteorological category which determine attenuation or Wind Wind direction in bins/ ° spee amplification Amet is given in [1,2]. d in bins Usually in projects the worst-case scenario is 60 120 150 180 210 240 270 300 330 at 0- 30- 90- - - - - - - - - - 29 59 119 10m 89 149 179 209 239 269 299 329 359 assumed for propagation in all direction and limit value for (m/s) the zone which is not in accordance with Croatian 0-1 19, 19,2 20 19,5 19,5 19,7 19,9 19,8 20 20,3 19,8 19,9 8 legislative (+1 dBA increase in residual noise or -5 dBA 1-2 22,4 21,5 20, 21,1 20,5 20,4 21 22,7* 23,5* 20,4* 23,4* 22,7* * * 7 from limit value for the area). 2-3 27,9 28,0 23, 27,6 26,9 24,8 28,5* 34,3* 41,0* 28,4* 27,1* 27,5* * * 4 3-4 31,1 - - 30,4 30,4* 28,8* - - - 42,2* 30,8* 30,8* * 3. MEASUREMENT AND MODELLING RESULTS 4-5 31,4 35,6 - - 35 34,4* - - - - 36,6* 33,7* * * 5-6 37,2 - - - 39,7* 37,5* - - - - 40,2* - * In this section the measured residual noise levels 6-7 - - - - 43,0* - - - - - - - and modelled levels from wind turbines are considered. 7-8 - - - - 43,7* - - - - - - - 8-9 - - - - - - - - - - - - 3.1 Residual noise levels *Small number of 10 min samples The residual noise levels are shown in Table 1. for two different positions during day, evening and night Table 1. Residual noise levels for MM1 and MM2 periods and the model for old wind turbines (all working positions during the night period (most critical) in maximum operating conditions). The residual noise levels at different bins for A problem arises when trying to collect the different positions are shown in table 1. at different time samples at higher wind speeds, so they are estimated by periods (day, evening, night). The residual noise levels are using polynomial fitting curve. mostly from old wind turbines installed before (without The levels of residual noise are shown in Fig. 2. this type of calculation). depending on wind speed at MM1 and MM2 (for all The purpose of this study was to see the directions and range of windspeed magnitudes at influence of new wind turbines on the present residual different day periods) and approximation fitting curve noise levels at different wind speeds and directions. with saturation at higher levels. The measurement place MM1 is under the influence of the old wind turbine noise so residual noise dependence on windspeed at 10m is Noise levels (dBA)-night Wind more representative for MM2. Wind direction in bins/ ° speed in bins 120 150 180 210 240 270 300 330 at 0- 30- 60- 90- - - - - - - - - 29 59 89 119 10m 149 179 209 239 269 299 329 359 (m/s) 0-1 23,3 23,4 23,0 23,1 23,5 23,0 23,2 23,1 23,1 23,2 23,2 23,3 * * * * * * 1-2 23,5 23,8 24,0 23,5 23,5 23,7 23,7 23,4 23,9 23,3 23,7 23,7 * * * 2-3 28,1 24,5 23,9 24,2 25,1 25,1 - - 25,5 24,4 25,4 25,2 * * Petošić et al.: MODELLING AND ASSESSMENT WIND TURBINE NOISE AT DIFFERENT METEOROLOGICAL CONDITIONS 86 AAAA – 2023 – IZOLA - Conference Proceedings The wind speed at hub height is estimated by using eq. 4. and it is used in calculation of sound pressure levels at different windspeeds (stability class and attenuation from wind speed direction in source receiver direction). h 𝑙𝑜𝑔𝑒[ +1] 𝑈(h) = 𝑈 𝑧0 0 ∙ (4) h 𝑙𝑜𝑔 0 𝑒[ +1] 𝑧0 where z0 is roughness parameter according HARMONOISE model (equation 5.90 in [2]) with roughness z 0=0.05 (comparison of measured velocities at different heights). The U0 (m/s) is wind speed velocity at h0=10m. 3.2 Modelling results for wind turbine noise Different testing scenarios (different wind turbine types and heights and their numbers) are tested to see the available working modes which would not exceed the level of residual noise by more than +1 dBA (a very strict rule for the situation with the sound powers at lower speed velocities). At lower speed velocities there is no possibilities to change working modes because the difference between modes is negligible. At higher wind speed velocities at 10m (>6m/s) there is a possibility to reduce the sound power (and electrical energy production significantly) however, the residual noise levels are significantly higher. The wind speed at critical locations is measured at 10 metres height and noise levels are measure at 4 metres on the border in the free field (having no influence of reflection from nearby objects). The ground factor has been tested for G=0 and G=0.2 and better agreement with some of the residual noise measurement results (downwind) was obtained with G=0.2 (old wind turbine results for samples where residual aerodynamic noise is estimated to be low). According to the ISO 9613-2:1997 the ground reflection factor should be G=0 however, the soil around wind turbines is not fully hard so G=0.2 is used (the difference between results is 0.8 dBA with the same sound power Fig. 2. Residual noise levels at position MM1 and MM2 with G=0 and G=0.2). In [1] it is written that the best due to wind speed magnitude at 10m solution is to put G=0 for the windfarm noise (the difference in the level is 0.8 dBA). It is visible that the residual noise increase versus There is some grass in the vicinity close to the wind speed at 10m according to saturation rule (averaged receiver so G=0.2 gives more realistic results rather than for all wind directions) is higher at position MM2 where assuming a stricter value G=0. The sound power levels of there is no influence of old wind turbines. wind turbines at different speeds and modes of working are shown in Fig. 3. Petošić et al.: MODELLING AND ASSESSMENT WIND TURBINE NOISE AT DIFFERENT METEOROLOGICAL CONDITIONS 87 AAAA – 2023 – IZOLA - Conference Proceedings 255◦ 29,1 16,7 30,3 23,7 285◦ 32,5 16,7 29,8 23,7 315◦ 33,5 16,7 28,6 23,7 345◦ 32,5 16,7 26,4 22,5 h=10m, v=3 m/s Res1 Res2 Res3 Res4 Direction 15◦ 24,2 20,8 31,1 29,9 45◦ 26,9 19,7 30,4 30 75◦ 30,6 23,7 30,4 30 105◦ 30,6 23,7 30,4 30 Fig.3. Typical sound power levels for wind turbine at 135◦ 30,6 23,7 30,4 30,1 different windspeeds at 10m height and different 165◦ 30,6 23,7 28,8 29,5 working modes 195◦ 30,6 23,7 36 30 It can be observed that at lower speeds there is 225◦ 30,6 23,7 38 31 no possibility to reduce sound power which is not the case 255◦ 28,9 20,2 40 30 with higher wind speeds. In the case of increasing modes, 285◦ 33,4 20,8 42,2 33,1 the production of electrical energy is significantly lower. 315◦ 33,1 21,8 30,8 30,8 The next step was to enter input parameters of 345◦ 32,8 23 30,8 29,4 sound sources in digital terrain model in mode 0 at *Results used for model calibration with measurement different speeds (arithmetic average of bin velocity) and results compare the results with measured residual noise at the same level. Table 2. Results for calibration of model with measured If the residual noise level is increased more than residual noise levels (old wind turbine) +1 dBA for the same wind direction and velocity (assumed 0 octas) and most unfavourable stability class during The results when all new wind turbines are different periods as obtained by measuring, the residual working during night period (0 octas cloudiness) are noise, the sound power od closest wind turbine is reduced shown in Table 3a) with increased values of residual noise if possible or turned off one by one until the limit in Table 3b). condition (+1 dBA total level of noise) is satisfied. The example for the results for wind speed at 10 Night, class E m around measurement locations and comparison h=10m, Pos1- Pos2- Pos3- Pos4- between old wind turbine levels with calibration point v=3 m/s new new new new Direction- MM1 is shown in Table 2. Results for new sources (at 15◦ 37,4 34,3 26,6 27,9 height h= 138 m) is shown in Tables 3a). 45◦ 37,4 34,2 27,4 27,8 75◦ 35,9 34,1 30,1 30,6 105◦ 34,9 32,2 32,3 33,7 Night, class E 135◦ 34,2 28,5 33,7 33,8 h=10m, v=4 m/s Pos1-old Pos2-old Pos3-old Pos4-old 165◦ 33,2 27,3 33,7 34,7 Direction- 15◦ 29,1 16,7 23,8 19,1 195◦ 30,6 27 33,7 34,9 45◦ 25,9 14,4 23,3 17,8 225◦ 30,5 27,4 33,5 34,9 75◦ 22,8 11,7 23,4 16,6 255◦ 34,3 28,2 32,6 34,1 105◦ 21,8 9,5 24,6 16,6 285◦ 35 32 29,5 31,5 135◦ 21,8 9,5 28,2 17,6 315◦ 35,9 32,8 27 31,4 165◦ 21,8 9,5 29,3 20,3 345◦ 36,5 34,3 26,4 28,8 195◦ 22,8 11,7 30,2 23,5 225◦ 26,8 15 30,3 23,7 Petošić et al.: MODELLING AND ASSESSMENT WIND TURBINE NOISE AT DIFFERENT METEOROLOGICAL CONDITIONS 88 AAAA – 2023 – IZOLA - Conference Proceedings Table 3a). The level of noise from new wind turbines (all working in mode 0) and increase in residual noise. A significant influence on the results at each location depending on the wind speed component in source-receiver direction is obvious. The difference in levels is up to 7 dBA when all wind turbines are turned on and working in am mode 0. h=10m, v=3 m/s Difference1 Difference2 Diff3 Diff4 Direction- 15◦ 13,4 13,7 1,3 2,1 45◦ 10,9 14,7 1,8 2,0 75◦ 6,4 10,8 2,9 3,3 105◦ 5,7 9,1 4,1 5,2 135◦ 5,2 6,0 5,0 5,2 165◦ 4,5 5,2 6,1 6,3 195◦ 3,0 5,0 2,0 6,1 225◦ 3,0 5,2 1,3 5,4 255◦ 6,5 8,6 0,7 5,5 285◦ 3,9 11,5 0,2 2,3 315◦ 4,6 11,3 1,5 3,3 Fig 4. Graphical representation of wind direction 345◦ 5,2 11,6 1,3 2,7 influence (same meteorological class E) on noise levels Red-increased the residual noise level more than +2 dBA around wind turbine sources at v=6m/s at 10m height for three wind directions a) 0◦, b) 90◦ and c) 180◦ Table 3b). Increased values of residual noise at location MM1, MM2, MM3 and MM4 when new sources are The final results for working modes at different turned-on periods and at different wind speeds are shown in Table 4. for middle wind direction in bin ( between 0◦ - 30◦ )(mid It is visible that at lower speeds the increase of 15◦). residual noise levels is significant and majority of wind turbines at these locations should be turned off which is not economic regarding the produced electrical energy. Angle (0-29)◦ The graphical representation of results with Wind speed at 10m _3 Day m/s 4 m/s 5 m/s 6 m/s three different wind directions (0◦, 90◦ and 180◦) and wind speed magnitude at 10m v=6m/s are shown in figures 4a) VA-V-I Mode0 Off Mode18 Mode18 ,b) c). VA-V-II Off Off Mode18 Off VA-V-III Off Off Off Mode18 VA-7 Off Off Off Off VA-9 Off Off Off Off VA-12 Off On(mode18) Off Off VA-11 Off Off Off Off VA-13 Off Off Off Off Angle (0-29)◦ Evening 3 4 5 6 VA-V-I Off Off Mode18 Off Petošić et al.: MODELLING AND ASSESSMENT WIND TURBINE NOISE AT DIFFERENT METEOROLOGICAL CONDITIONS 89 AAAA – 2023 – IZOLA - Conference Proceedings residual noise is measured). In simulations upper value of VA-V-II Off Off Off Off the windspeed magnitude bin was used to estimate sound VA-V-III Off Off Off Off pressure level and mid direction in bin to compare the VA-7 Off Off Off Off simulation results from new sources to the residual noise VA-9 Off Off Off Off levels with the same input parameters (wind direction, VA-12 Off Off Off Off magnitude and cloudiness). In further steps more VA-11 Off Off Off Off comparison between modelling and measurement results VA-13 Off Off Off Mode18 will be done in order to assess the modelling uncertainty. Night(23:00-7:00) 5. REFERENCES Angle 0-29 3 4 5 6 [1.] Hansen C.H, C.J. Doolan, K.Hansen. Wind farm VA-V-I Off Off Mode18 Mode18 noise:Measurement assessment and control, Wiley, VA-V-II Off Off Off Mode18 2017. VA-V-III Off Off Off Mode18 [2.] Bies A.D, C.C Hansen, C.Q., Howard, Engineering Noise Control, 5th Edition, CRC Press 2018. VA-7 Off Off Off Off [3.] Croatian Legislative regarded to noise, VA-9 Off Off Off Off https://zakon.hr/z/125/Zakon-o-za%C5%A1titi- VA-12 Off Off Off Off od-buke, 2020 VA-11 Off Off Off Off [4.] Croatian legislative for limit noise values, Pravilnik o VA-13 Off Off Off Off najvišim dopuštenim razinama buke s obzirom na vrstu izvora buke, vrijeme i mjesto nastanka (nn.hr), Table 4. Final results with working modes of wind 2021. turbines at different wind velocity magnitudes at 10m for [5.] ISO 1996-2:2017, Acoustics - Description, different periods of day (stability classes) and wind measurement and assessment of environmental noise -- Part 2: Determination of sound pressure directions levels, International Organization for Standardization. It is evident that at lower wind speed in this [6.] ISO 1996-1:2017, Acoustics - Description, configuration with proposed distances (i.e., min 800m) measurement and assessment of environmental the wind turbines should stay turned off at lower wind noise --Part 1: Basic quantities and assessment velocity magnitudes during the night which is not procedures, International Organization for Standardization. profitable. [7.] ISO 9613-2:1996 Acoustics — Attenuation of sound during propagation outdoors — Part 2: General 4. CONCLUSION method of calculation, International Organization for Standardization. A lot of parameters (mainly meteorological and [8.] Joint Research Centre, Institute for Health and propertied of terrain) have an influence on determination Consumer Protection, Anfosso-Lédée, F., Paviotti, M., of wind turbine modes, noise abatement and production Kephalopoulos, S., Common noise assessment methods in Europe (CNOSSOS-EU) : to be used by the of electrical energy if the wind turbines are installed very EU Member States for strategic noise mapping close to residential objects (>=800m). The critical following adoption as specified in the Environmental parameters are ground reflection factor G, estimation of Noise Directive 2002/49/EC, Publications Office, wind speed and velocity at critical locations and where the 2012, https://data.europa.eu/doi/10.2788/32029 wind turbine hub is located (possible different wind directions near residential objects and hub). In this paper the direction of wind is assumed to be the same at hub and at measured locations (where Petošić et al.: MODELLING AND ASSESSMENT WIND TURBINE NOISE AT DIFFERENT METEOROLOGICAL CONDITIONS 90 Evaluation of model-based noise protection study based on in-situ vibroacoustic railway track analysis Ivo Haladin, Krešimir Burnać University of Zagreb, Faculty of Civil Engineering, Department for Transportation Engineering Abstract: In the design phase, newly built or reconstructed railway infrastructure requires a noise protection study to investigate noise sources and the need for noise protection around the railway line. Usually, such studies are based on modelling noise sources and their propagation through the environment, and they tend to overestimate actual noise levels. Therefore, in certain cases, before noise protection is built, investors want to re-evaluate the position, height, and length of noise protection walls foreseen by the noise study. This paper presents a novel approach to the evaluation of a noise protection study on a newly reconstructed railway line in Croatia. It involves vibroacoustic techniques for the characterization of vehicle pass-by noise and railway track properties such as rail roughness and track decay rate. Based on such properties and a series of in-situ measurements on a reference railway track and the observed newly built track, the noise protection study has been re-evaluated and need for noise protection walls reduced. Results of noise propagation models have further been tested and verified when the railway line has been completed. Keywords: Railway noise, pass-by noise measurements, noise propagation model, vibroacoustic measurements, track decay rate, rail roughness 1. INTRODUCTION insufficient when assessing the noise levels from railway traffic in locations where traffic conditions are highly A complex project of a railway infrastructure takes variable and have a higher influence on the noise levels significant amount of time from the preliminary design (i.e. railway stations). In those situations, it is advised to phase, through numerous environmental studies, main conduct additional in-situ railway noise measurements to design phase, tendering, building the railway get more suitable results [1]. infrastructure and finally getting a running permit and Moreover, input in this study (done 10 years prior to starting rail operations. In Croatian engineering practice construction) anticipated use of mostly DMU passenger this process can take from 5 up to 15 years. trainsets. It was later decided that all the passenger trains It is therefore a challenging task to comply with all current running on Zaprešić-Zabok section will in fact be new EMU standards and regulations and to keep up with other trainsets (Končar HŽ-6112 series) which are, of course developments around the railway infrastructure which much quieter compared to old DMUs. interconnected with it. Project required control measurements of environmental In this case, in 2022 a project of Reconstruction and noise prior to building foreseen noise protection walls to electrification of railway line Zaprešić – Zabok has been check if the walls are needed in the current in-situ near the end of construction phase. The noise protection conditions with new EMU trains running at design speed study from 2013 has foreseen protection by means of 19 of up to 120 km/h. Since the section Zaprešić - Zabok had noise protection walls, total length of 1000 m. Noise levels jet to go through final inspection and get a running permit, have been estimated using a noise prediction model that the speed was reduced to 40-80 km/h on the section and is based on the German method “Schall 03”. This method electrification has still not been in operation. has shown some excellent results for railway noise It was therefore impossible to perform in-situ calculation because it contains some parameters that environmental noise measurements with proposed have a considerable effect on noise levels (track operations (EMU trains running at up to 120 km/h). curvature, vehicle length). However, they could be Haladin et al.: Evaluation of model based noise protection study based on in-situ vibroacoustic railway track analysis 91 AAAA – 2023 – IZOLA - Conference Proceedings This posed a challenge for the engineering team hence 3) 24-hour environmental noise measurements – in-situ they turned to Faculty of Civil Engineering University of measurements that were done on the test section. In the Zagreb for a solution proposal. The proposed method was total noise levels, to validate whether considerable noise protection from 4) post processing - DMU trains operating on the tested train traffic is required by reviewing the noise protection section at speed of before the electrification were study and forecasting future noise levels. The method replaced with EMU trains from the referent section. consisted set of noise measurements a reference railway section (M101 State border – Savski Marof – Zagreb main Only after the modernization and electrification were station), and a tested section (R201 Zaprešić – Zabok). It finished, under realistic circumstances, could actual noise included measuring 24h environmental noise levels at 19 levels be tested (design speed, number of trains, traction locations along the test track. Actual train pass-bys though type - new, quieter EMUs). had to be determined at a referent track where the trains EU regulation no.1304/2014 [3] prescribes the standard could run at the speed of 120 km/h. To successfully HRN EN ISO 3095:2013 [2] for the determination of the transfer train pass by noise from the referent location to maximum noise levels from a train pass-by. Other actual locations on test track, vibroacoustic parameters of standards that are used to determine track condition are the track had to be determine because at these speeds, HRN EN 15610:2019 [4] which is used for acoustic the dominant source of railway noise is interaction roughness measurements of the rail running surface, and between rail and wheel. HRN EN 15461:2011 [5] which gives guidelines for After the vibroacoustic parameters were determined and measurements of vibration damping of the track evaluated, it was possible to apply the measured noise structure. Vibro-acoustic properties of the track (acoustic levels from the reference track (120 km/h), as relevant on rail roughness and track decay rate (vibration damping of the tested track. Doing this, a necessary condition for both the track)), are an important part of the pass-by noise tracks to meet the vibroacoustic parameters according to measurements because, with their determination and the standard [2] is met, and they can be declared evaluation, two different railway lines can be defined as comparable in acoustic terms. comparable [6]. With the determination of those two To determine the functional relationships of noise levels parameters, it is possible to determine the share of the at different speeds of pass-by and distance from the track track contribution to total noise levels during the train to apply the measured pass-by levels from the referent passage. If the results are similar for referent and test track to various speed/distance conditions at the tested track, the pass by at a referent track can be regarded valid railway track, three regression curves were created. This also for the test track. allowed a translation of pass-by noise levels of new EMU vehicles at speeds of up to 120 km/h from referent track 2.1. VIBRO-ACOUSTIC MEASUREMENTS to each of 19 locations on tested track. To validate the results, a comparison based on pass-by The following measurements were carried out to inspect speed, noise levels, and distance from the track was made the acoustic resemblance of the test railway track R201 both for reference and tested track, for different running Zaprešić – Zabok and the referent railway section of the conditions (acceleration and deceleration). track M101: visual inspection of the track superstructure, determination of the track decay rate according to HRN 2. MEASUREMENTS EN 15461:2011 [5], and measurements of the acoustic railhead roughness according to HRN EN 15610:2009 [4]. To get the 24-hour environmental noise levels in the 19 Track decay rate is a vibroacoustic property of the railway locations where noise barriers are planned to be track that describes how vibrations propagate along the constructed based on the noise study, several rail, or what is the capacity of the track to absorb the measurements and analysis steps needed to be vibrations [7]. The goal of this measurement was to conducted. 1) Vibro-acoustic measurements – to determine the establish the quality of the referent and tested rail track acoustic similarity between referent and test track in terms of noise and vibration emission from the track- section. side. Several parameters can influence the track decay 2) Pass-by measurements - to determine the noise levels rate, such as rail cross-section, rail pad stiffness, rail from the passing railway vehicles operating at different inclination, and distance between two sleepers, and those speeds and different distances from the rail track axis at parameters should be uniform for both railway sections referent track. according to the standard [4]. Measurements are conducted using a modal hammer with appropriate Haladin et al.: Evaluation of model based noise protection study based on in-situ vibroacoustic railway track analysis 92 AAAA – 2023 – IZOLA - Conference Proceedings stiffness (containing an accelerometer), and with reduced speed V= 80 km/h because of a nearby accelerometers installed on the head of the rail in vertical railway station. and horizontal directions. Measurement devices that were used Brüel & Kjær - Direct acoustic roughness measurement of the railhead Hand-held Analyzers Type 2260, 2250, 2245, and 2270. running surface is a method for the determination of Microphones were installed at distances 7.5m, 11.5 , 15 railhead roughness, which generates noise levels during a m, 19 m, 25 m and 29 m at each measuring location. vehicle pass-by [8]. As a part of this research, acoustic Pass-by measurements have been repeated on the tested railhead roughness was measured on one location on the track after the tested track got its usage permit (Vmax = 120 referent track M101 and 3 locations on the tested track km/h) to evaluate and confirm the method that was used R201. It was measured using a RAILPROF 1000 device for using the same described procedure. direct rail roughness measurement, with a 1m effective measurement length. 2.3 24 HOUR ENVIRONMENTAL NOISE MEASUREMENTS The tested track on the railway section R201 is acoustically comparable to the referent track M101, according to the To inspect if the number of locations and length of the barriers from the noise study (based on a calculation results of track decay rates and acoustic rail roughness. method with generic parameters) is in correspondence to Furthermore, noise measurements on the referent track the noise levels on the railway line, 24-hour in-situ can be used to determine noise levels on the tested track. environmental noise measurements were conducted in The abovementioned measurements are explained in accordance with ISO 1996[9]. Measurements were more detail in the paper from the same authors [6]. conducted using Brüel & Kjær - Hand-held Analyzers Type 2260, 2250, 2245, and 2270, with measuring points at 4.0 2.2. PASS-BY NOISE MEASUREMENTS m above the ground. Distance from the microphone to the track varied from 11 to 36 meters, dependent on field To measure the noise levels emitted by passing EMU 6112 conditions and accessibility. traveling at various speeds and distances from the rail 24-hour environmental noise levels that were recorded on track axis, pass-by measurements were made, both on the tested track section, were recorded while DMU – HŽ referent and tested railway sections all according to 7121 trains were in operation, which were operating at standard EN ISO 3095 [2]. Pass-by measurements on the 85 Substitution of 7121 referent track were made at two locations, the first one in 80 train with 6112 train in a time interval an open track section with maximum speed V 75 max = 120 ] km/h, and the second one 70 20 Pa 65 re 85 B d 60 [ s =1t, 55 80 eqLA 50 7121 75 45 6112 70 ] 40 Pa 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 20 65 Time [s] reB [ds 60 , t=1 eqLA 55 50 45 40 :00 :00 :00 :00 :00 07:00 08:00 09:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19 20:00 21:00 22 23 00:00 01:00 02:00 03:00 04 05:00 06 07:00 Time [hh:mm:ss] Fig. 1. Example of 24-hour environmental noise record with the pass-by noise levels from 7121 (DMU) train (blue), and 6112 (EMU) train (red), and an example of their substitution (detail) Haladin et al.: Evaluation of model based noise protection study based on in-situ vibroacoustic railway track analysis 93 AAAA – 2023 – IZOLA - Conference Proceedings Function of distance and noise level HŽ 6112 - REF lower speeds (Vmax=80 km/h). Those operating 85.0 conditions would not fit the future situation after the modernization and electrification, and because of that, 80.0 ] y = -5.15ln(x) + 91.03 Pa noise records of those DMUs had to be substituted with R² = 0.8247 20 re 75.0 noise levels of new EMU-s that will be running after the B d[l e reconstruction end (Fig. 1). vle 70.0 e Pass-by noise levels for EMU trains were recorded on the is y = -6.931ln(x) + 94.904 oN R² = 0.8363 referent railway section, where trains are operating at 120 65.0 km/h. ACC DEC Log. (ACC) Log. (DEC) 60.0 7.5 10.0 12.5 15.0 17.5 20.0 22.5 25.0 27.5 30.0 Distance [m] 3. RESULTS ANALYSIS Fig. 2. Function of distance and noise level for EMU train 6112 on referent track (acceleration and deceleration) Based on conducted measured quantities described in chapter 2 a methodology for calculating environmental Following that, two regression formulas were extracted noise levels on the tested track, based on the in-situ based on the curves above (for acceleration and measurements of pass-by noise that were conducted on deceleration): the referent track is explained in more detail. Furthermore, a comparison between pass-by noise levels, Lp,V0,D = -5.15*ln(D) + 91.03 (R^2=0.82) (acceleration) (2) calculated based on a methodology explained in Chapter Lp,V0,D = -6.931*ln(D) + 94.904 (R^2=0.84)(deceleration) (3) 3.1, and measured pass-by noise levels that were where: measured at the tested track (validation measurements) L was made. p, V0, D noise level from train pass–by (in dB) at the reference speed V0 and distance D D chosen distance from the rail track axis [m] 3.1 REGRESSION CURVES After pass-by noise levels were recorded on the referent 3.2 VALIDATION OF THE METHOD track, a general equation for calculating the equivalent To validate the method described in 3.1, a comparison noise levels (depending on the different speeds of the between measured values (L,meas) of pass-by noise levels vehicle) was used [3] (with different n-coefficient for on the tested section after the modernization (Vmax = 120 acceleration and deceleration) (1). km/h), and calculated values (L, calc) based on the Lp,V = Lp,V0+n*log (v/v0) [dB] (1) measurements conducted on the referent section (Vmax = where: 120 km/h), normalized at V = 100 km/h, and calculated Lp, V noise level from train pass–by at speed V according to the beforementioned method with V arbitrary speed for which noise level is regression curves. determined Lp, V0 noise level from train pass–by at reference speed Regression curves (100 km/h) The regression curves shown in Error! Reference source V0 reference train pass – by speed (100 km /h) not found. and Error! Reference source not found. , were n coefficient (25 for acceleration, 20 for calculated for both acceleration and deceleration. They deceleration) are based on the results gathered from the referent track The n-coefficient from the formula (1) was determined empirically based on curve fitting to actual measured as well as the validation results from measurements pass-bys. Regression curves were created for the variation carried out on the tested track, after the opening of the of distances from the rail track axis, based on the electrified line. When the results of the referent and equivalent noise levels normalized to a reference speed of tested tracks are shown on a graph, they have an excellent 100 km/h (Lp, V0) (Fig. 2). Regression curves are based on match in the trendlines for both acceleration and results of pass-by noise on the referent section. deceleration. Haladin et al.: Evaluation of model based noise protection study based on in-situ vibroacoustic railway track analysis 94 AAAA – 2023 – IZOLA - Conference Proceedings these noise levels is on the safe side for further Function of distance and noise level HŽ 6112 - Acceleration 85.0 calculations on 24 noise levels based on modeled pass-by noise levels. ] 80.0 Pa 20 reB Table 1. Comparison of noise levels (L, meas) from in-situ 75.0 vel [d pass-by measurements on the test track and calculated e leis 70.0 o noise levels based on the regression curves (L, calc) N 65.0 TEST REF Log. (TEST) Log. (REF) V Acc (<) L,meas L, calc Δ Lm - 60.0 (meas) D [m] 7.5 10.0 12.5 15.0 17.5 20.0 22.5 25.0 27.5 30.0 [dB] [dB] Lc [dB] Distance [m] [km/h] Dec (>) 7.5 m 81.6 80.9 0.7 Fig. 3. Function of distance and noise level for EMU train 102 < 15 m 76.3 77.3 -1.0 6112 on the referent track and tested track – 25 m 72.0 74.7 -2.7 acceleration 7.5 m 80.7 81.0 -0.3 101 > 15 m 76.4 76.2 0.2 Function of distance and noise level HŽ 6112 - Deceleration 25 m 72.5 72.7 -0.2 85.0 7.5 m 76.9 76.0 0.9 80.0 ] 65 < 15 m 71.7 72.4 -0.7 Pa 20 25 m 68.1 69.8 -1.7 re 75.0 B 7.5 m 75.0 73.6 1.4 vel [d 70.0 e leis 52 < 15 m 71.1 70.0 1.1 oN 65.0 25 m 67.8 67.4 0.4 TEST REF Log. (TEST) Log. (REF) 7.5 m 74.5 73.2 1.3 60.0 7.5 10.0 12.5 15.0 17.5 20.0 22.5 25.0 27.5 30.0 41 > 15 m 68.4 68.4 0.0 Distance [m] 25 m 65.2 64.8 0.4 Fig. 4. Function of distance and noise level for EMU train 7.5 m 73.7 73.2 0.5 6112 on the referent track and tested track–deceleration 41 > 15 m 68.3 68.4 -0.1 25 m 64.8 64.8 0.0 Pass-by noise levels Calculated pass-by noise values from referent track are compared to measured noise levels for different actual 4. FINAL RESULTS AND DISCUSSION pass-bys on the test track after opening of the line, for To calculate the 24 hour pass-by noise levels for operating different operating speeds, type of operation condition with EMU trains and Vmax of 120 km/h, (acceleration or deceleration), at three different distances from the axis along with the difference between L, substitution of DMV trains in the 24hour signal has been meas, and L, calc noise levels, Table 1. For most of the situations, conducted according to Fig.1 for all of 19 positions where the difference between measured and calculated values is noise protection walls are foreseen by the noise below or around 1 dB, with a few exceptions where protection study. With newly calculated noise levels, a calculated values are higher than the measured ones (Δ > final recommendation could be made on the need for 1 dB): noise protection walls . Based on pass-by measurements - first pass-by at 102 km/h (acc) and difference in noise levels Δ = -2.7 dB on the referent and tested track, a significant reduction in - second pass-by at 65 km/h (acc) with Δ = -1.7 dB the number of locations and length of noise barriers was at the distance D = 25m from the axis. made. The noise study recommendation was to build a The cause for higher calculated values could be variations total number of 19 noise barriers (1000 meters in length), in environmental conditions such as wind, humidity, and the results from in-situ measurements have proven background noise, etc which are most evident at the that only 5 noise barriers (240 meters in length) needed furthest distance from the (25 m from the track) for the to be constructed ( Fig. 5). referent and tested section. The model overestimation of Haladin et al.: Evaluation of model based noise protection study based on in-situ vibroacoustic railway track analysis 95 AAAA – 2023 – IZOLA - Conference Proceedings Noise study In-situ measurements 19 Noise barriers 5 Noise barriers Fig. 5. Number of noise barriers that are required for railway noise protection, based on the noise study (left), and based on in-situ measurements (right) 5. CONCLUSION verify the need for noise protection by means of in-situ Noise protection study results rely on a noise propagation 24-hour control measurements of environmental noise, model calculated to a certain accuracy for an area of has been made to validate the need for noise protection interest. Accuracy of end results and relies heavily on walls on a newly reconstructed and electrified railway line noise source definition in the model, calculation method Zaprešić – Zabok. used and data validation at the end. Due to lengthy Since the operating conditions could not be met prior to process of railway infrastructure project implementation, running permit for the track (electrification and Vmax of some of conditions and regulations related to noise levels 120 km/h), to achieve representative 24h environmental can change and therefore challenge the conclusions of the noise levels, a substitution of train noise signals has been noise protection study. Therefore, a correct decision, to proposed, based on pass-by noise measurements on a Haladin et al.: Evaluation of model based noise protection study based on in-situ vibroacoustic railway track analysis 96 AAAA – 2023 – IZOLA - Conference Proceedings referent track with similar operating conditions. To verify https://doi.org/10.1016/j.apacoust.2018.03.006, the acoustic similarity of the tracks, vibroacoustic 2018. [2] European Commission, Commission Regulation (EU) parameters of rail surface roughness and track decay rate No. 1304/2014 of 26 November 2014 on the have been determined and compared. technical specification of interoperability relating to Regression curves have been established to calculate the subsystem “rolling stock - noise.”, Official noise levels for any distance from track/train speed Journal of the European Union, 2014 (17), 2014. combination that is needed for the noise study. [3] CEN, HRN EN ISO 3095:2013 Acoustics - Railway Validation of the model based on actual measured values applications - Measurement of noise emitted by after track opening has been performed. < 1 dB difference railbound vehicles, 2013. [4] CEN, HRN EN 15610:2009 Railway applications -- between calculated noise levels on the referent track and Noise emission – Rail roughness measurement recorded in-situ noise levels on the tested track (Table 1) related to rolling noise generation, 2019. can be observed, and it was concluded that the model is [5] CEN, HRN EN 15461:2011 Railway applications - well built and could be further used. Noise emission - Characterisation of the dynamic Based on proposed procedure, a revaluation of noise properties of track sections for pass-by noise protection resulted in reduction of needed noise measurements, 2011. protection from initial 1000 m to 240 m in length. This [6] Haladin, I., Burnać, K., Koščak, J. Vibroacoustic track analysis and noise measurements on the R201 significant reduction greatly impacted the project budget Zaprešić-Zabok railway line, Road and Rail and timeline resulting in timely issuing of running permit Infrastructure VII, Proceedings of CETRA 2022, 2022. and start of operations. [7] Haladin I, Lakušić S, Košćak J. Measuring vibration damping level on conventional rail track structures, Gradjevinar 2016;68. ACKNOWLEGDMENTS https://doi.org/10.14256/JCE.1532.2015. [8] Lakušić S, Haladin I, Jukić A, Andraši N, Piplica P. Rail This paper was written as a part of the Croatian science roughness measurement and analysis in frame of foundation project “HRZZ - Young Researchers’ Career rail vehicle pass-by noise measurements. Road and Development Project – Development of DIV Elastic Rail Rail Infrastructure II, Lakušić S, editor., Zagreb: Fastening” DOK-2021-02-9981. Sveučilište u Zagrebu Građevinski fakultet; 2012. [9] CEN, ISO 1996-1:2016 Acoustics — Description, Measurement and Assessment of Environmental 6. LITERATURE Noise — Part 1: Basic Quantities and Assessment Procedures, 2016. [1] Džambas, T., Lakušić, S., Dragčević, V., Traffic noise [10] CEN, ISO 1996-2: 2017 Acoustics -- Description, analysis in railway station zones, Applied measurement and assessment of environmental Acoustics;137:27–32., noise -- Part 2: Determination of sound pressure levels , 2017. Haladin et al.: Evaluation of model based noise protection study based on in-situ vibroacoustic railway track analysis 97 STEEL REILWAY BRIDGE NOISE, LACK OF REDUCTION EFFECT ON AIRBORNE NOISE DUE TO VIBRATION DAMPERS POSSIBLY ACTING AS NOISE SOURCES Rok Rudolf Slovenian National Building and Infrastructure and Civil Engineering Institute Laboratory for Thermal Performance and Acoustics Dimičeva ulica 12, SI-1000 Ljubljana Abstract: Steel railway bridges can be quite loud, especially in the low frequency range, and they typically last over 100 years- presenting a very persistent noise pollution problem. One way to tackle this problem is to use noise dampers on the bridge itself, damping vibrations and reducing the airborne noise emissions from the bridge. In one practical application case however, it became apparent that the noise dampers on a specific pilot bridge have substantially reduced the vibration of the bridge, but the airborne noise was barely affected. One possible explanation for this is that the noise dampers themselves act as sound sources. To research their contribution to total sound levels of the bridge, we have mounted one of the noise dampers on a vibration machine in a laboratory, and measured emitted sound during various excitation schemes. This way, we were able to assess the noise emitted by the damper at vibration levels typically encountered on a bridge during individual train pass-byes. This allowed us to add dampers as noise sources in a computer noise emission model of the bridge. Using this model, we assessed the effect that dampers as noise sources have on total noise, as measured on site Keywords: steel railway bridges, vibration dampers, noise sources, noise emission model 1. INTRODUCTION The system for monitoring showed a peculiar set of result on a pilot bridge however. Once vibration dampers Steel railway bridges represent a significant noise source, were installed, vibration levels on the bridge web plates especially at low frequency ranges, producing a were substantially reduced in the 1/3rd octave frequency “thundering noise” sensation to nearby listeners. Steel bands with middle frequencies of 50 Hz and 63 Hz, where bridges tend to vibrate at those ranges, producing large dampers were designed to be most effective[4]. noise emissions that are difficult to reduce with Simultaneously, noise emissions from the bridge were conventional noise control measures like noise also measured, characterized by bridge correction factor barriers[1]. Due to long life-span of steel railway bridges - (BCF) parameter: that is, a difference in Z-weighted sound typically over 100 years- this problem will not go away any pressure levels measured at the bridge, and at some time soon. distant reference site on the same track, where there is no A noise and vibration monitoring system for steel railway influence of bridge noise emissions. This way, the noise bridges was developed as part of the EU Horizon 2020 emissions particular to the bridge itself can be shown project Assets4Rail (part of Shift2Rail) and tested on a separately from the noise emissions from the railway pilot bridge with and without specifically designed tracks away from the bridge. However, results showed vibration dampers used as noise control [2]. These that BCF before and after installation of vibration dampers dampers were installed on the main supporting element – has changed much less than expected in the 1/3rd octave bridge web-plates[3]. frequency band at 63 Hz where BCF was highest[4]. 1st Author Surname et al.: Paper title 98 AAAA – 2023 – IZOLA - Conference Proceedings and oriented in a way that would reduce noise emissions on a level plain (e.g., walking plates). Since no explanation was found with existing bridge elements, I have turned to check if a new element could be the source of additional noise emissions, that could explain the lack of reduction in BCF of the pilot bridge. Namely could the vibration dampers themselves, that vibrate at this frequency range by design, radiate enough sound to reduce the effect on the BCF. Could dampers themselves be a relevant sound source? To test this hypothesis, one noise vibration damper was mounted on an axial vibration table MTS Bionix 370, and shaken at various frequencies with varying force, including the settings that correspond to the vibration excitation of the actual bridge under traffic (averaged over all trains) at main resonance frequency 53 Hz. Next, the sound emission from this system was measured in the room where vibration table is located, in effect treating this room as an reverberation room, so sound power levels of each damper could be determined according to a standardized method ISO 3743-2 [5]; once other contributions like background noise due to machine operation were discounted. And finally, a computer model of noise emission of the bridge in question was prepared in Predictor-LIMA software, according to the EU noise Fig.1. substantial difference (>20 dB) in web plate assessment methods CNOSSOS-EU [6].Noise sources vibration due to damper installation (top graph – corresponding in power to noise dampers assessed in the German: „Schnellepegel“) does not translate to reduction previous step were added to the model to calculate their in noise emissions, characterized by BCF (bottom graph– contribution with measured total noise levels at the German: „Brückenzschlag“). Both shown before (“von bridge [4] acting as reference. The process explained in Einbau”) and after installation (“nach Einbau”). Source: this article can then give us an answer of how much of the TUB [4] total bridge noise is in fact due to noise vibration dampers acting as noise emission sources at relevant low So, as can be seen from figure 1, despite substantial frequencies (around 53 Hz) [4]. reduction in the vibration of the bridge, bridge noise characterized by the BCF parameter remained only minimally effected. 2. MEASUREMENT SETUP How could this be? First, it was speculated that another part of the bridge was also vibrating and emitting sound Vibration damper is a central metal frame with 24 metal in the relevant frequency range around 63 Hz. However, sandwich rods sticking out on two sides, in two rows. Each no obvious bridge element could be found, that could act rod is of a sandwich construction with a bitumen-based as this additional source. Vibrations and subsequent layer between two steel layers. This bitumen-based emissions at these frequencies would imply a large middle layer ensures inner damping of each rod, while the source, comparable to the main web plates of the bridge length and dimensions of steel layers determines the that were the main focus of the investigations. No such resonant frequency, allowing for damper “tunning” – by elements existed on the bridge in question (see also next making the rods longer or shorter, the resonance point for bridge description). Any elements considered frequency where the damper is most effective is lowered were either much too small (e.g., hand rails), or too light or raised. This particular damper was tunned to the 1st Author Surname et al.: Paper title 99 AAAA – 2023 – IZOLA - Conference Proceedings resonance frequency of 53 Hz, same as the dominant same time, sound levels were recorded inside the room resonance frequency of the bridge web-plates, and with a B&K sound level meter with a microphone mounted weighted in at 33 kg. The damper was designed and on a rotating boom, in accordance with standard ISO produced by Assets4Rail partner Schrey & Veit GmbH, 3743-2[5]. Background noise levels in the room Lbck [dB] based on the noise and vibration measurements at the while machine was operating but not actively vibrating, as bridge [4]. well as room’s reverberation time RT60 [s] were also The vibration damper was mounted on an axial vibration measured, to be used to complete the calculations table MTS Bionix 370 in a concrete walled room of volume according to the same standard. 121,4 m3 at Slovenian National Building and Civil Engineering Institute (from Slovene: ZAG). The mounting to the vibration piston was done by 4 screws in the same 3. VIBRATION DAMPERS AS SOUND SOURCES way the damper would be mounted on an actual bridge. The results are presented on the following figure, with parameters LZeq measured at various 1/3rd octave bands with a rotating boom. For comparison the excitation of the actual bridge at damper locations before vibration damper installation is included. This excitation, corresponding to an average regional train pass-by, was measured at future damper location1 as maximum velocity of 𝑣𝑚𝑎𝑥 = 8,12 𝑚𝑚/𝑠 with a peak at 53 Hz[7]. From this we can get to maximum displacement with: | 𝑣 𝐴| = 𝑚𝑎𝑥 = 24,4 𝜇𝑚 (1) 2𝜋𝜈 Fig.2. vibration dampers from the measured bridge, as received This displacement is marked as “BV” (Bridge Vibration) line on figures. 75 BV 70 B][d 65 z H 60 50 55 at 50 L Zeq 45 10 20 30 40 50 60 70 80 90 100 110 Fig.3. dampers mounted on the vibration table by screws, Displacement A [µm] note the 12 rods in two rows to one side in the picture Exict. @ 26,5 Hz Exict. @ 40 Hz Exict. @ 53 Hz (res.) Exict. @ 63 Hz foreground Backg. N. (50 Hz) The vibration damper was then vibrated with forced Fig.4. graph of measured sound pressure levels at 1/3rd oscillation at various frequencies ν [Hz] and various forces octave band with middle frequency 50 Hz, re. maximum – characterized by the maximum displacement in a displacement, at various frequencies of excitation. vibration cycle A [µm] setting to the vibration table. At the Logarithmic trend lines (dashed). 1 Dampers were located at two points on each web-plates, where vibration measurements confirmed the largest maximum velocity (displacement). 100 AAAA – 2023 – IZOLA - Conference Proceedings Where: As can be seen from figure 3, excitation at frequencies - 𝐿𝑝 are mean band pressure levels measured with other then expected damper resonance frequency yields rotary boom around the reverberation room (LZeq low noise in 50 Hz 1/3rd octave band – at levels around at middle frequency 50 Hz for our case), background noise. This is not the case at resonance - 𝑇𝑛𝑜𝑚 is the nominal reverberation time, frequency at 53 Hz and at 26,5 Hz (half of resonance determined by centering the measured frequency), although at excitation levels encountered on reverberation time of the room, normalized to the actual bridge (marked “Bridge Vibration”) the noise RT60 at 1000 Hz acc. to standard [5], 1,1 s in our case, levels are barely above background noise. - 𝑉 is the room volume, 121,4 m3 in our case, Another interesting aspect can be seen if we look at - constants of 𝑇0 = 1 𝑠, 𝑉0 = 1 𝑚3, and correction sound pressure levels in other 1/3rd octave bands that of - 13 dB. include excitation frequencies, as presented on the next To determine the sound power at measured bridge graph. excitation levels (|𝐴| = 24,4 𝜇𝑚 @ 53 Hz), we first determine the pressure levels. Our measurement is very 65 close to the background sound levels at that excitation, as BV B] 63 we measured a value 1,1 dB above background (see figure 61 y [d 3). Rather than try to deduct the background value, we 59 enc 57 extrapolate the expected pressure level from noise levels 55 at higher excitation (larger displacement). With equation ct. frequ 53 xi for trend line for excitation at 53 Hz from figure 3: e 51 at 49 47 L Zeq 𝐿𝑝 = 𝐿𝑍𝑒𝑞@50 𝐻𝑧 = 7,3 ln(𝐴) + 27,912 (3) 45 10 20 30 40 50 60 70 80 90 100 110 Displacement A [µm] , we get the mean pressure level and by putting this into equation (2), we get to the value of emitted sound power Exict. @ 26,5 Hz Exict. @ 53 Hz (res.) Exict. @ 63 Hz Backg. N. (25 Hz) of each individual damper, when vibrated with measured Backg. N. (50 Hz) Backg. N. (63 Hz) bridge excitation levels: 𝐿𝑤 = 40 𝑑𝐵, at 1/3rd octave band with middle frequency 50 Hz. Fig.5. graph of measured sound pressure levels in 1/3rd This sound power can then be used as an input parameter octave bands that include excitation frequency – relation of the noise emission model of the bridge, to see the to maximum displacement. Background noise in 3 expected contribution of our sound source (vibration relevant bands also presented. dampers) to total noise levels. As can be seen from figure 4, the noise measured is highest near expected resonant frequency. Measuring in 4. NOISE EMISSION MODEL bands near 53 Hz still shows increased noise with respect to background – note that 1/3rd octave band with middle Noise emission model was prepared, using Predictor-Lima frequency 63 Hz includes frequencies from 56,2 to software (ver. 12.01) for environmental noise. The model 70,8 Hz. At frequencies further away (e.g., 26,5 Hz), any included vibration dampers as point sound sources. It also increase is negligible compared to background noise. included microphone measuring position as was used Once sound pressure levels are determined, they can be during the measurement campaign in the Assets4Rail used to determine band sound power according to ISO project [4],[7]. This measurement position was chosen 3743-2 method [5], with the following equation: according to the measurement system [2] based on the ISO 3095 standard [8] – measuring was done 1/3rd from 𝑇𝑛𝑜𝑚 𝑉 the north-east side of the bridge, at 7,5 m from the middle 𝐿𝑊 = 𝐿𝑝 − 10 log + 10 log − 13 𝑑𝐵 (2) 𝑇 of the track and 1,2 m above top of the rail head (ToR). 0 𝑉0 1st Author Surname et al.: Paper title 101 AAAA – 2023 – IZOLA - Conference Proceedings the model that extra noise emissions due to dampers acting as sound sources amount to 𝐿𝑐𝑎𝑙𝑐,𝑒𝑞,50𝐻𝑧 = 41 𝑑𝐵 in 1/3rd octave band at 50 Hz, calculated at position of measurement at the bridge. One can compare this calculated value to average total noise level measured at this position near bridge after noise damper installation (𝐿𝑚𝑒𝑎𝑠,𝑒𝑞,50𝐻𝑧 = 88 𝑑𝐵) for regional trains [7]. With noise levels contribution calculated to 47 dB below measured total noise levels, it becomes quite clear that vibration dampers do not constitute any relevant contribution to total noise, even at noise bands where they are arguably the loudest. Fig.6. photograph of the bridge Source: TUB [7] 5. CONCLUSION As can be seen from our model results, we are no closer to finding the missing airborne noise source of a particular steel railway bridge, measured in Assets4Rail program. The vibration of web-plates was damped, but the noise persists. It might be possible that the result could be improved upon with some more advanced calculation model, that might for instance include resonance effects of the space below bridge itself. However, even if that were the case it seems impossible that this could amplify the noise of vibration dampers by nearly 50 dB. So, even Fig.7. photograph of the bridge between web-plates, with with our simpler emission model, it can conclusively be vibration dampers installed – 2 per each web-plate said that vibration dampers are a negligible noise source Source: TUB [7] compared to the noise of a steel railway bridge. The results of sound power of such dampers, as measured and calculated in chapter 3, might be more pertinent if such dampers were used in a less noisy environment. 6. REFERENCES [1] D. Thompson, “Chapter 11 - Bridge Noise,” in Railway Noise and Vibration, D. Thompson, Ed., Oxford: Elsevier, 2009, pp. 359–397. doi: https://doi.org/10.1016/B978-0-08-045147- Fig.8. Predictor-Lima noise emission model 3.00011-6. [2] J. Boehm, C. Gramowski, M. Ralbovsky, L. Rizzeto, and R. Rudolf, “Assets4Rail Deliverable D1.2: Measurement results at the bridge were used to calibrate Report on a noise emission monitoring solution the model – spectral emissions from sources were varied for steel railway bridges,” 2021. Accessed: Aug. until model values were equal to values measured before 23, 2023. [Online]. Available: vibration dampers were installed. http://www.assets4rail.eu/wp- After, vibration dampers were included in the model with content/uploads/2021/12/A4R-D1.2.pdf sound power as derived in chapter 3, it was calculated by 1st Author Surname et al.: Paper title 102 AAAA – 2023 – IZOLA - Conference Proceedings [3] C. Gramowski, “Assets4Rail Deliverable D4.2: and Consumer Protection., Common noise Report on improved work approach for set up of assessment methods in Europe (CNOSSOS-EU) : bridge dampers” (non public) Berlin, 2020. to be used by the EU Member States for strategic [4] C. Gramowski, T. Hanisch, M. Ralbovsky, and R. noise mapping following adoption as specified in Rudolf, “Brückenakustik: Fortschritte bei the Environmental Noise Directive 2002/49/EC. Analyse, Theorie und Minderungsmaẞnahmen OPEU, 2012. im Rahmen des Shift2rail-projektes Assets4rail,” [7] T. Hanisch, M. Ralbovsky, C. Gramowski, and R. in BAHNAKUSTIK 15. – 16. November 2021 Rudolf, “Assets4Rail Deliverable D1.2.1: Bridge Planegg München , München, 2021. Noise Measurements in Pressig, Germany, [5] SIST EN ISO 3743-2:2019 Acoustics - before and after the installation of bridge noise Determination of sound power levels of noise dampers,” (non public), 2022. sources using sound pressure - Engineering [8] SIST EN ISO 3095:2013 Acoustics - Railway methods for small, movable sources in applications - Measurement of noise emitted by reverberant fields - Part 2: Methods for special railbound vehicles. 2013. reverberation test rooms. 2019. [6] Stylianos. Kephalopoulos, Marco. Paviotti, Fabienne. Anfosso-Lédée, and Institute for Health 103 Psychoaoustics of Pseudosound in Turbulent flow of Centrifugal Fan used in Household Appliance Jurij Prezelj, Jurij Gostiša University of Ljubljana, Faculty of mechanical engineering Abstract: As the performance metrics of household appliances from various manufacturers converge, acoustic performance is emerging as a pivotal differentiator for consumers. While strides have been made in noise control, further reductions in noise levels are challenging due to inherent physical constraints. Hence, the focus is shifting from merely reducing noise levels to reshaping noise to be less intrusive or even pleasant. While noise arising from vibrations has been largely manageable, aerodynamically generated noise, primarily due to the chaotic nature of turbulent flow, presents a significant challenge. The early turbulent flow studies introduced the term "pseudosound," highlighting the intricate connection between turbulence and acoustics. This research, with a focus on the centrifugal fan in a tumble dryer, employs a psychoacoustic approach to analyze turbulent flow. Our findings indicate a correlation between psychoacoustic features, namely roughness and fluctuation strength, and the turbulent properties of flow when analyzed through hot wire velocity signals. This leads to a compelling question: Why does human auditory perception possess the ability to discern the turbulent characteristics of flow? Keywords: Psychoacoustics, pseudo sound, aeroacoustics, turbulence, centrifugal fan 1. INTRODUCTION profile, and cost-effectiveness, ensures their widespread use in both domestic and industrial settings. Forward-curved-centrifugal (FCC) fans are distinct for This ubiquity has piqued the interest of numerous two primary reasons: they possess a significant rotor research groups, prompting investigations to address and outlet to inlet ratio (around 3:1) and are equipped with enhance the primary shortcoming of FCC fans: efficiency. numerous short, chorded blades (typically around 40). Furthermore, noise control pertaining to these fans Given their unique design, they've acquired various names continues to be a focal point of scientific exploration, as worldwide: drum-rotor fan in Europe, squirrel-cage fan evidenced by numerous studies [2], [3], [4], [5]. overseas, and also referred to as Sirocco fan or multivane impeller, a nomenclature credited to Eck [1]. 1.1 Motivation and problem formulation Owing to their compact size and low noise emission, FCC fans have become a staple in both industrial and Historically, the suboptimal efficiency of forward- residential HVAC systems. They are particularly favored in curved fans has been largely attributed to flow separation applications demanding high airflow at moderate within the interblade channel on the shroud side. This pressures, especially when efficiency is not the top vortical flow spans approximately one-third of the rotor's priority. Frequently, the shape of the fan's inlet channel is width, leading to the choking of the interblade channel dictated by spatial constraints, leading to suboptimal and, consequently, a reduction in efficiency. Many aerodynamics and, consequently, diminished efficiency. authors have delved deep into this phenomenon, yet Yet, in spite of their limited efficiency, their utility in studies addressing the effects of inlet flow channel design delivering high flow rates with a moderate pressure boost, on fan performance have been sparse. coupled with their compactness, agreeable auditory In his seminal work, Eck [1] dedicated an entire chapter to forward-curved fans, highlighting the flow 104 AAAA – 2023 – IZOLA - Conference Proceedings separation zone at the fan's inlet. While he proposed fluctuating velocity is responsible for these acoustic several design guidelines, he underscored the existing disturbances. Lighthill's analogy [19] casts light upon this gaps in foundational knowledge for designing and concept by introducing an equivalent source term in the computations. Since Eck's insights, numerous research wave equation. Specifically, turbulence acts as a weakly groups have ventured into the realm of forward-curved radiating quadrupole source, the strength of which fans. For instance, Kind [6], [7] elucidated the intricate depends on the fluctuating hydrodynamic velocity. flow patterns within the rotor, emphasizing the significant Notably, acoustic perturbations are substantially weaker axial and circumferential nonuniformities, particularly in magnitude than their hydrodynamic counterparts. evident at flow rates beneath the best efficiency point It's pivotal to recognize that pressure fluctuations are (BEP). Tsutomu's tripartite study [8], [9] on blade design not solely the offspring of acoustic modes. Vortical and casing presented both numerical and experimental fluctuations also play a role in generating these data, culminating in insights on optimal blade numbers, fluctuations. There's a fundamental difference in angles, and volute dimensions. propagation speeds between acoustic waves and Subsequent research from the Oviedo University team hydrodynamic pressure fluctuations. The former travel at in Spain [10], [11] focused on the aerodynamic and the speed of sound relative to the fluid, while the latter acoustic properties of forward-curved fans, especially the move with the local fluid velocity. As such, it's imperative dual inlet variants utilized in automotive HVAC systems. to differentiate between the two. Acoustic fluctuations Their work highlighted the pivotal role of volute tongue give rise to what we perceive as sound, whereas design in noise reduction and how the operating point hydrodynamic fluctuations lead to pseudo-sound, which is profoundly impacts flow through the rotor. Parallel to this, also discernible to an observer [19]. Montazerin's group at Amirkabir University of Technology In the realm of subsonic speeds, the pseudo-sound undertook comprehensive research, culminating in a 2016 field holds dominance within and proximal to turbulence, publication [12] that covered key aspects such as inlet constituting the acoustic near field. As one ventures configuration and rotor/volute design, substantiated by further away, this field is overshadowed by the acoustic both experimental and numerical data. radiation field, which diminishes more leisurely with Echoing Montazerin's focus, Pham Ngoc Son [13] distance. Intriguingly, the pseudo-sound field closely explored the influence of inlet geometry on performance mirrors the characteristics of the pressure field in metrics. Notably, their studies centered on fans with a incompressible flows. Dominated more by inertial forces free inlet configuration, where the inlet remains exposed than by compressional effects, it lacks wave propagation. to the ambient environment. As their results indicated, This unique nature has earned it the moniker the relatively minor portion of the inlet channel doesn't "pseudosound" [20]. considerably affect fan characteristics. Contrarily, To break it down mathematically, considering the Golamian [14] evaluated flow straighteners at the inlet, turbulent flow's random nature, we can assume different such as tubes and zig-zag plates, discovering their segments of the flow act as uncorrelated or incoherent detrimental impact on pressure head and efficiency. sources. This allows us to express the total pressure, ptot, Addressing spatial constraints, Xuanfeng Wen et al.'s work as: [15] explored the repercussions of a discontinuous volute p = p + p + p (1) tot 0 PS A profile, revealing its role in both decreasing efficiency and amplifying noise. Here, pPS symbolizes the RMS value of hydrodynamic The intertwining of aero-acoustically generated noise with pressure oscillations due to turbulence, commonly psychoacoustic features has garnered attention, termed pseudosound. On the other hand, pA stands for especially in the automotive sector [16], [17]. Controlling the RMS value of pressure oscillations stemming from flow-induced noise is paramount in product development, acoustic propagation. Though p0 represents static necessitating the consideration of near acoustic fields in pressure, it's typically excluded from considerations when the sound spectrum induced by flow [18]. focusing on dynamic phenomena in frequency ranges exceeding 1 Hz. It's essential to note that the RMS values 1.2 Pressure oscillations in turbulent flow as pseudo- of these oscillations are distance-dependent, with their sound origin being the source. The decay patterns are distinctive: while pseudosound's RMS decays following an inverse Turbulent flows are characterized by changes in their third-power law with distance, acoustic pressure's RMS pressure and velocity fields. Such alterations naturally adheres to an inverse law. This contrast is vividly depicted lead to acoustic perturbations. In essence, hydrodynamic in Figure 1. 105 AAAA – 2023 – IZOLA - Conference Proceedings radiative contribution, often termed as pseudosound in literature [20], [24], relates to the traversal of eddy structures in the flow. Consequently, it travels at a pace considerably slower than sonic velocity, especially at lower Mach numbers. Much of the scholarly research has concentrated on deriving the pressure field from velocity measurements in turbulent flows using hot-wire techniques [25], [26], [27], [28]. Obi derived the pressure field from velocity measurements using Euler's equation of fluid motion (referenced as Eq.1) and opted to ignore the unsteady term [29]. Data secured via PIV underscored that the notable turbulence intensity observed between bodies Figure 1: Sound pressure (red) and pseudosound originates from the oscillatory motion tied to vortex pressure (blue) as function of the distance r shedding. The impact of vortex motion is evident as a robust correlation between velocity and pressure gradient Within turbulent flows, pressure fluctuations are [29]. Here, nn and ss stand as orthogonal vectors, with nn predominantly driven by the inertial effects linked to the being normal to the surface. vortices, such that ptot≈pPS>>pA. Outside these flows, pressure fluctuations are largely due to propagating p  t ( )  u  t ( )  u  t ( )  u  t ( ) n n n (2) waves, signified by p = − − u t ( ) + u t ( ) tot≈pA. Although turbulence-induced n  t n  n s  s  acoustic waves tend to have smaller amplitudes, they exhibit a more gradual decay [21]. An examination of the topology in the kx–ω spectrum Suzuki introduced a diagnostic method aimed at of each reconstructed pressure field highlighted the identifying instability waves in a subsonic round jet, existence of three modal classes: hydrodynamic, acoustic, employing a phased microphone array. This detection and a hybrid known as hydro-acoustic. The hydrodynamic algorithm draws parallels with the beamforming and acoustic modes were defined by a spectral energy technique, which is typically harnessed in conjunction bump with phase velocities roughly equivalent to jet with a far-field microphone array to pinpoint noise origins. speed and at least equal to the ambient sound speed, Statistical evaluations of the results align with the concept respectively. Hybrid modes showcased multiple energy that the pressure field corresponds to instability waves bumps, each having distinct phase velocities [22]. developing within the turbulent mean flow [21]. Interestingly, the potential energy of waves, By utilizing the wavenumber-frequency spectrum represented by E method, one can assess levels of both acoustic and pp(k) spectrum, can significantly surpass their kinetic energy, as described by E turbulent flow pressure fluctuations, in addition to dd(k) spectrum. This disparity even leads to questions about the very notion of estimating convective velocities present in the near acoustic waves. Such observations prompted numerous acoustic field. Notably, acoustic pressure fluctuations scholars, including aeroacoustics pioneers like Lighthill, to have been found to dominate over those of turbulent flow refer to this phenomenon as the pseudosound regime, at elevated frequencies within the near acoustic field [18]. distinguishing it from genuine acoustics [19]. It's essential When positioning a microphone within an airflow, it to highlight that pseudosound tends to dissipate in higher becomes susceptible to the turbulence present in said Mach number regimes, with its observations typically flow. Moreover, any turbulence generated due to the valid up to M<0.1M<0.1 [19]. microphone's presence further interacts with its Felli posited that intermittent pressure peaks, diaphragm. This interaction between turbulence and the triggered by the transit of eddy structures, can be diaphragm prompts the microphone to register noise perceived as pseudosound. In contrast, consistent levels attributable to this interference, rather than purely background fluctuations might be understood as sound capturing the acoustic wave. For airflow velocities [23]. A notable challenge with near-field pressure approaching 27m/s, foam windscreens seem ill-suited for measurements is that only a fraction of the energy linked acoustic measurements, given their inability to shield with pressure fluctuations radiates as sound. The effectively against these turbulent disruptions [30]. remaining fluctuations, which don't align with the linear wave equation, can't be identified as sound. This non- 106 AAAA – 2023 – IZOLA - Conference Proceedings 1.3. Rotational sound of Aerodynamic Origin configuration, temperature vacillations, and so on. However, first-order estimates can be derived from The total noise level of a centrifugal fan is ussulay empirical data obtained for the most part from dominated by discrete frequency tones at the blade experimentation in the aviation industry. The earliest passage frequency and its higher harmonics. This is a measurements of jet noise demonstrated that intensity result of the rotor blade interaction and the guide vanes and noise power varied very closely with the eight power [31, 32, 41]. Many of the noise sources are near the of the jet exit (Lighthill’s eight power law), and it is now trailing edge of the rotor blades [33]. Fluctuating loads at generally agreed that the overall sound power can be the tips of the blades generate the dominant noise. Rotor expressed with quadrupoles as [38]: blades are typical dipole sources [34]. Rotational noise of aerodynamic origin at the design 8 2  v D 0 P = K (1) point of operation ( Q q q des,  p des) is the result of pressure 5 c 0 fluctuations caused by periodic fluid forces. Thrust and drag forces are induced on the blades as they move Where K q is a quadrupole radiation constant through the air. Rotational noise is also generated by describing nozzle configuration, turbulence impulsive interaction of the rotor blades with the inflow distortion and nearby stationary obstacles, such as the characteristics, temperature etc., D is the jet diameter, diffuser vanes and return passages. Aerodynamically and 0 is the density of ambient air, [38]. generated noise from rotational origins at off-design An interaction of the turbulent airflow with the solid operation and at constant rotational speed is theoretically surface also generates a broadband noise, which has no equal to that at the design point of operation. But the correlation with a rotation frequency. Curle [39] showed rotational noise changes due to varying of both, the that the sound generated by turbulent flow in the region steady and unsteady loading effects, which characterize of a solid body is exactly analogous to the sound radiated the off-design operation [35]. It was also observed that by a 'free space' distribution of quadrupoles sources plus the number of blades does not significantly affect the a surface distribution of dipoles. Sound power of acoustic noise level at the BPF [36]. radiation from an acoustically compact solid in turbulent flow can be expressed as [39]: 1. 4 Non-rotational sound of Aerodynamic Origins 6 2  v L Non-rotational noise of aerodynamic origins is 0 P = K (2) d d 3 c generated by random forces at the design and off-design 0 point of operation. These forces are induced by the non- uniform flow fields, by turbulence and by interaction of where K d is a dipole radiation constant describing the the turbulent flow with the rigid structure along the air shape of the solid body, turbulence characteristics, flow through the electric motor. Air flows through the temperature, etc., and L is a typical length of the solid suction unit to cool the electric motor. Heated air, exits body of the solid object. Accordingly, these dipoles should the suction unit with high velocity, forming a turbulent air be more efficient generators of sound than the jet. There is a large disparity between the energy of the quadrupoles of Lighthill’s theory if the Mach number is flow in the non–linear field and the acoustic energy in the small enough. far field. The hydrodynamic pressure p ranges from 104 Pa The non-rotational noise can also be generated by the to 106 Pa, whereas the acoustic pressure p’ ranges from rotating stall and by the surge [3]. Acoustic radiation can 10−5 to 10 Pa. Therefore, the radiated acoustic energy of an unsteady flow is a very small fraction of the total be attributed directly to the pressure oscillations due to energy in the flow. In general, the total radiated sound unsteady operating conditions of the suction unit can be power of a turbulent jet scales with  ( v 8/ c 5 described as, [40, 41]: 0 ), and for a dipole source arising from pressure fluctuations on 4 2  v d 0 P = K (3) surfaces inside the flow scales with  ( v 6/ c 3 m m 0 ), where v c 0 denotes the characteristic flow velocity and c 0 the speed where K m is a monopole radiation constant describing of sound [37]. suction unit geometry, and d is the diameter of the The nature of the noise from jets cannot be accurately opening from the suction unit. predicted, owing to the complex nature of the jet itself and the uncertainties associated with turbulence, nozzle 107 AAAA – 2023 – IZOLA - Conference Proceedings At higher flow rates, towards free delivery ( v >> v des) noise for example, and in the calculation of an unbiased aerodynamic non-rotational turbulent noise is generated annoyance metric. due to higher jet flow velocities, i.e., boundary layer A definition of the loudness of tones can be vortex shedding and incidence declination of the flow on constructed from the results of experiments such as the suction side of the rotor blade. At lower flow rates ( v loudness ‘magnitude estimation’. The ‘loudness level’ of a << v des), the non-rotational aerodynamically generated sound (introduced in the twenties by Barkhausen) is noise is caused by vortices, flow declination, and defined as 'the sound pressure level of a 1 kHz tone in a especially due to the onset of the rotating stall and surge plane wave and frontal incident that is as loud as the phenomena. Combined sound power generated by sound; its unit is “phon”.’ (Zwicker & Fastl 1990). So a monopole, dipole and quadrupole sources can be written sound that is as loud as a 1kHz tone with a sound pressure as: level of 40dB (for example) is said to have a loudness level of 40 phon. This principle can be used to define the 8 2 6 2 4 2  v D  v L  v d loudness of tones by comparing them with an equivalently 0 0 0 P = K + K + K (4) c q 5 d 3 c c m c loud 1kHz tone. 0 0 0 Reference: 1000 Hz tones. The loudness level of a tone of 1000 Hz in phones L is per definition its SPL (dB). According to the ISO definition, the loudness of this 1000- Hz tone in sones N' is found by: L −40 SPL 10 N '= 2 (5) This means that per definition the loudness of a 40-dB SPL, 1000-Hz tone is 1 (sone). Rule of thumb: From equation (1), it follows that a 10 dB increase in loudness level means a doubling in loudness. This holds quite well for 301  b 16 ark 0,35  N  For L = 40 (7) N  + 00 , 0 05  sone  LN<1 Where only for critical-band rates larger than 16 Bark does the factor increase from unity to a value of four at So, while it is more usual in acoustics to see the the end of the critical-band rate near 24 Bark. This "loudness" of a signal expressed in dB(A), a better considers that sharpness of narrow band noises increases measure of the perceived loudness can be found by unexpectedly strongly at high centre frequency. proper application of the critical bandwidths - A specific The booming index can then be calculated by loudness can be calculated from the dB level for each third weighting these low frequency elements using the octave band using the assumption that' a relative change following weighting function and comparing them with in loudness is proportional to a relative change in the loudness in the rest of the spectrum: intensity.' (Zwicker and Fastl 1990) So, values of specific loudness (N') in sone per Bark can 1.4.3 FLUCTUATION STRENGTH be calculated using a power law. Masking curves can then Owing to modulated sounds can elicit two different be constructed around these levels representing the kinds of hearing sensations: at low modulation effect of critical bands. The final value for loudness (N) is frequencies up to a modulation frequency about 20Hz, the then calculated as the integral (i.e., the area) under the hearing sensation of fluctuation strength (F) is produced. curve and is presented in sones. At higher modulation frequencies, the hearing of bark 24 roughness (R) occurs. From modulation frequencies N =  N' dz (8) around 20Hz, there is a transition between the hearing 0 sensation of fluctuation strength and that of roughness. A 0,23  0,23  model of fluctuation strength based on the temporal  QTQ   1 E  N'= 08 , 0  +  −  (9)   1 variation of the masking pattern can be expressed as:  E E D    2 2   TQ   L  F = (12) In which, QTQ is the excitation at threshold in quiet f 4 mod + and ED is the excitation that corresponding to the 4 f mod reference intensity I0=10-12 W/m2. which shows the relationship between fluctuation 1.4.2 SHARPNESS strength (F) and the masking depth (∆L) of the temporal Sharpness is the ratio of high frequency level to overall masking pattern, as well as the modulation frequency level. It is calculated as the integration of specific loudness (fmod). which exhibits the distribution of loudness across the critical bands multiplied by a weighting function, divided 1.4.4 ROUGHNESS by total loudness (hence, sharpness is level-independent). To describe roughness quantitatively, three Here the zwicker’s method is used for the sharpness S parameters are used. For amplitude modulation (AM), the calculation: important parameters are the degree of modulation and b 24 ark g( z)  z N'( z dz   ) modulation frequency. For frequency modulation, it is the (10) S = 0 11 , 0 frequency modulation index and modulation frequency. b 24 ark N'( z dz  ) Using the boundary condition that a 1-kHz tone at 60 dB 0 and 100%, 70Hz AM, produces the roughness of 1 Asper, the roughness (R) of any sound can be calculated by: where the denominator gives the total loudness (N). The upper integral is like the first moment of specific b 24 ark R = 3 , 0 f mod L ( z) dz (13) loudness over critical-band rate, but uses an additional  E 0 109 AAAA – 2023 – IZOLA - Conference Proceedings integrating psychoacoustic metrics into velocity signal where fmod is modulation frequency, and ∆LE is the analysis. masking depth in critical band. The critical bands are 4. Interrelation of Sound and Turbulent Flow: There auditory band pass filters, and its band number scale exists a pronounced linkage between sound and turbulent flow. Drawing from this relationship, we have postulated (“psychoacoustic” or “Bark” scale) is a frequency scale of that methodologies used in sound signal processing can the numbered critical bands 1 through 24, named Bark also serve to analyze turbulent flow. More explicitly, (named after von Barkhausen), and the width of a given strategies formulated for the psychoacoustical critical band is approximately: 100 Hz, at center interpretation of sound can be repurposed as descriptors frequencies below 500 Hz; 0.2xfc, at center frequencies of turbulent flow characteristics. above 500 Hz. These bands derive from the frequency-to- place transform on the basilar membrane. 2. EXSPERIMENT 1.5 Conclusions based on literature survay and Experiments were conducted on a measurement setup scientific contribution. constructed in accordance with ISO 5801 standards, specifically the installation of category B, which entails a 1. Flow Separation and Efficiency: A significant factor free inlet and a ducted outlet. A schematic representation affecting the reduced efficiency of the forward-curved fan is flow separation. This phenomenon begins at the inlet, is provided in Figure 3 and 4. The setup includes: progresses through the blades, and extends into the 1. Test Section with a Spiral Casing (Figure 2-1): The volute. Such vortical flow accounts for approximately one- fan, housed within a spiral casing, serves as the primary third of the rotor width, thereby leading to diminished focus of the experiment. The design of this casing was flow rates and efficiency. influenced by guidelines put forth by Montazerin [12]. A 2. Psychoacoustic Metrics in Turbomachinery Noise differential pressure transducer, designated for capturing Evaluation: In recent times, psychoacoustic metrics have the pressure difference across the fan, was connected at emerged as a preferred method for assessing point A (as shown in Figure 2-1), with its secondary port turbomachinery noise. Intriguingly, our survey did not exposed to ambient conditions. yield examples where these metrics were applied to non- 2. Fan Specifications: The runner used had a width of acoustic signal analysis. There's a tangible relationship 60 mm, a diameter of 155 mm, and featured 40 blades. between pressure and velocity fluctuations in fluid flows, Powering the fan was a 400W servo motor, ensuring a which validates the application of psychoacoustic metrics consistent rotation frequency of 50 Hz irrespective of the to velocity signal analysis. Consequently, metrics such as loudness (DIN 45631/A1), sharpness (DIN 45692), operating conditions and facilitating shaft torque fluctuation strength, and roughness have been employed measurements. to elucidate flow properties that impact efficiency and 3. Flow Rate Measurement (Figure 2-2): An ISO 5167- noise levels. Based on these insights, an innovative compliant orifice with flange tapings was integrated to prototype of the inlet channel was designed, enhancing measure the flow rate. the fan's efficiency, and minimizing noise emissions. 4. Operating Point Modulation (Figure 2-3): An 3. Influence of Inlet Configuration on Performance: auxiliary fan combined with a motorized throttle valve The focal point of the research presented in this paper was used to create a diverse set of operating points. centers on the impact of inlet configuration design on 6. Noise Measurement: Noise profiles were captured local inlet flow field attributes and, consequently, the using a Norsonic Nor140 Sound Analyser placed 1 meter performance dynamics of the FC fan. We undertook a from the fan. series of experiments to gauge the fan’s performance The initial segment of the experiment was devoted to traits, assess the inlet flow field using HW anemometry, gauging the performance and noise attributes associated and measure emitted acoustic pressure fluctuations. Average velocities were computed for each data point and with different inlet configurations: the free inlet, the pre- visualized on a contour graph relative to the given volume existing inlet channel (referred to as the TD inlet), and a flow rate. Meanwhile, sound pressure signals were newly devised inlet channel. For the subsequent phase: depicted on a spectrogram, contingent on volume flow 7. 3-Axis Traversing System: To identify deterministic rate. An avant-garde approach was introduced by flow patterns within the inlet channel, a system utilizing three motorized linear rails driven by stepper motors was 110 AAAA – 2023 – IZOLA - Conference Proceedings crafted. A Dantec type 55P11 1D hot-wire anemometer, 4. RESULTS placed orthogonal to the flow direction, surveyed 144 distinct points as depicted in Figure 2 (right section). Prior The results present a comparative analysis of the free to data collection, the probe underwent calibration. At inlet, existing inlet, and the newly designed inlet every measurement location, a 10-second velocity signal configurations. The initial segment evaluates integral snapshot was collected at a sampling frequency of 25 kHz. aerodynamic and acoustic parameters, specifically This data collection was facilitated by an NI9222 module, performance and noise characteristics. Meanwhile, the housed within an NI9147 cDAQ chassis. Automation was a subsequent segment delves into the local properties of key feature, with both measurement and probe the inlet flow. positioning being controlled by a custom-built software developed in the NI Labview environment. 4.1 Integral aerodynamic and acoustic properties 8. Inlet Design Exploration: Alongside the existing inlet (shown in Figure 3, left), a novel inlet was Figure 5 depicts the performance characteristics for conceptualized, derived from observed flow patterns the free inlet and both ducted inlet configurations. Each within the inlet channel. The research process configuration reveals a notable unstable operating region encompassed multiple design iterations. Initially, the up to approximately 200 m3/h. Beyond these flowrates, team aimed to emulate the free inlet configuration. fan operation stabilizes across all inlet configurations. The However, this design, when juxtaposed against the pre- occurrence of partially filled interblade channels, which existing inlet channel, exhibited reduced efficiency. leads to an increase in pressure for FC fans compared to Consequently, a secondary approach was undertaken. fans with backward-inclined blades at very low flow rates, This new design, as presented in Figure 3 (right), was is comprehensively discussed by Eck [1 - p.114]. A fitted optimized to guide fluid streamlines towards the inlet second-order polynomial curve aptly describes the stable opening, culminating in superior efficiency and diminished segment of these characteristics. The Best Efficiency Point noise levels. (BEP) is found at 390 m3/h for the free inlet, 300 m3/h for the existing inlet, and 340 m3/h for the new inlet channel. The hydraulic efficiency at BEP is approximately 50% for the free inlet, 37.5% for the existing inlet, and 41.5% for the new inlet channel. Regarding noise characteristics, as illustrated in Figure 6, all inlet configurations produce lower noise levels at reduced flow rates. However, the lowest noise levels do not align with the BEP as suggested by fan noise theory. Figure 3: Progression of the inlet channel design - For the free inlet, the Leq(A) levels tend to be higher Original version (left) compared with the redesigned compared to the two ducted configurations. This can be outline that follows fluid streamlines (right). ascribed to the inlet channel's role in noise dampening. Thus, a direct comparison of these cases may not be appropriate. A comparison between the existing and new inlet channels reveals a 2 dB(A) noise reduction at higher flow rates with the new inlet design. In contrast, differences below 250 m3/h appear minimal. This noise disparity arises from the altered aerodynamics, as everything else, barring the inlet channel geometry, remained consistent. With the new inlet channel, the Figure 4: Depiction of the hot-wire anemometer observed noise reduction and associated efficiency probe, part of a 3-axis traversing system (marked as 1), improvement are credited to decreased vorticity and and a microphone aligned with the fan's axis (denoted as enhanced flow conditions within the inlet channel. 2). 111 AAAA – 2023 – IZOLA - Conference Proceedings vortex (Figure 7 right C). The vortex in region C is believed 700 70 to be influenced by regions A and B, which themselves 600 60 a] originate from the rotor's nature and the spiral casing. The 500 50 y [%] e [P photo depicted in Figure 7 left was taken with the existing 400 40 reas inlet operating at 200 m3/h. Due to the inherent e inc 300 30 constraints of smoke visualization. The limitations of lic efficienc ur au 200 Free inlet 20 smoke visualization technique – dilution of smoke at high ress ydr P Existing inlet H flow rates and view obstructed by the channel walls, make 100 10 New inlet comparison of phenomena with respect to flow rate or 0 0 inlet channel variation inconvenient. Hence the results of 0 200 400 600 Volume flow rate [m3/h] hot wire anemometry were used. Figure 5: Performance and efficiency characteristics of the fan in free inlet configuration (black), with existing (red) and new inlet (right). 77 76 Free inlet 75 Existing inlet 74 B] New inlet 73 [d 72 (A) Figure 7: Visualization of the flow field in the existing 71 70 inlet channel at 200 m3/h (left) and a scheme of flow Leq 69 field as interpreted based on the visualization (right). 68 67 50 100 150 200 250 300 350 400 450 500 Volumetric flow rate [m3/h] Figure 6: Noise characteristics of the fan in free inlet configuration (black), with existing (red) and new inlet (right). While the maximal efficiency benchmark set by the free inlet remains about 10% superior, the new inlet channel design marks significant progress. To understand the underlying mechanisms that boost efficiency, further investigation was undertaken, the results of which are detailed in the subsequent subsection. Figure 8: Normalised velocity standard deviation and 4.2 Local flow properties at the inlet plane psychoacoustic metric levels distribution over the inlet area as a function of flow rate in the case of free inlet Local flow patterns at the fan inlet were identified through a combination of smoke visualization techniques and laser To reveal the flow dynamics in regions A, B, and C, we sheet illumination, supplemented later by hot wire applied psychoacoustic metrics to the velocity fluctuation anemometry. From the visualization, three distinct signals. Figure 8 displays the average velocity magnitude, regions emerged: a well-populated region with minimal turbulence intensity, and four psychoacoustic metrics— vorticity (Figure 7 lright A), a turbulent eddy zone with loudness, sharpness, fluctuation strength, and limited throughflow (Figure 7 right B), and an roughness—as functions of the flow rate. The results are intermediate region marked by a pronounced continuous 112 AAAA – 2023 – IZOLA - Conference Proceedings visualized as contour plots (with consistent scale ranges, embodies a hiss-like quality of sound, dominated by high- barring the velocity distribution) illustrating each frequency energy. Elevated sharpness levels, indicative of variable's distribution across the inlet area on fgures 8 to dominant high-frequency fluctuations, are typically 10. The red-marked region represents the existing inlet, observed in the turbulent eddy region. This pattern shares while the blue indicates the new inlet channel. a similar spatial distribution to that of loudness. Reviewing the velocity distribution, two prominent Consequently, it can be inferred that turbulent regions zones can be identified: 1. a high-velocity zone on the inlet predominantly exhibit high-frequency fluctuations. opening's left side and 2. a low-velocity pocket on the right, positioned behind the volute tongue. This observation aligns with earlier conclusions drawn from the visualization data. The high-velocity zone mirrors the well-populated area with minimal vorticity, while the low- velocity segment matches the turbulent eddy zone with sparse throughflow. Notably, the latter is also characterized by intense turbulence, confirming earlier assumptions. Using loudness as an analytical metric offers deeper insights into fluctuation behaviors within the turbulent zone, essentially reflecting the level of velocity fluctuation. An evident decrease in both loudness and sharpness is evident in the area labeled A in Figure 7. This change is attributed to the narrowing of the new channel at that specific location (as seen in Figure 3 and4). It seems this contraction positively directs the flow, diminishing vorticity within the turbulent eddy region (Figure 7). Figure 10: Local flow properties in the plane at the inlet of the fan for existing (red) and new inlet channel (blue) with respect to the volumetric flow rate. The region situated between the well-filled area and the turbulent eddy region—where a distinct, continuous vortex is evident (as shown in Figure 7, right C)—displays heightened values across all psychoacoustic metrics, pointing to a broadband fluctuation characteristic. Uniquely, this is the sole region demonstrating elevated roughness and fluctuation strength. Meanwhile, the flow dynamics throughout the remainder of the inlet region are characterized by either high or low frequencies. The similar patterns exhibited by roughness and fluctuation strength suggest loudness modulations at frequencies both below and above 30 Hz. From an aerodynamics Figure 9: Normalised velocity standard deviation and perspective, we associate the zones with increased psychoacoustic metric levels distribution over the inlet fluctuation strength and roughness to the presence of area as a function of flow rate in the case of TD inlet. extensive vortices. Such large-scale vorticity arises due to the pronounced velocity gradient in the transitional zone While loudness does not provide specific between the well-filled region with minimal vorticity (as information about the frequency band of the fluctuation, seen in Figure 7, right A) and the turbulent eddy region the addition of sharpness, roughness, and fluctuation (Figure 7, right B). Notably, at flow rates of 350 and 500 strength offers a more comprehensive picture. Sharpness m3/h, a slight decrease in fluctuation strength is evident 113 AAAA – 2023 – IZOLA - Conference Proceedings with the new inlet (Figure 7, right B), potentially explaining 4. Innovative Psychoacoustic Metrics: The study the earlier mentioned efficiency improvements and noise introduces a novel methodology by employing reductions. psychoacoustic metrics on velocity fluctuation signals. At 100 m3/h, heightened values of roughness and This method not only expanded our understanding of fluctuation strength manifest in the predominantly well- airflow dynamics but also proved invaluable for flow filled area with minimal vorticity—the left segment of the pattern interpretation, highlighting dominant inlet opening. This suggests the presence of large-scale fluctuation frequency bands relative to positions on vorticity characteristic of low flow rates, which impedes the inlet plane. the inlet channel, subsequently impacting efficiency. 5. Efficiency Factors: The redesigned inlet showcased a On a broader note, the lowest noise levels notable reduction in high-frequency fluctuation, correspond with a flow rate of 200 m3/h, a condition specifically in the turbulent eddy region, as well as a marked by diminished values across all psychoacoustic decrease in vortex presence. These improvements metrics. This relationship underscores the correlation directly contributed to the observed 4% efficiency between fluctuation intensity and noise emission. When spike and 2 dB(A) noise reduction. considering volume flow rate, notable metric distribution In summary, the blend of traditional methodologies with changes were observed at 100, 200, and 350 m3/h. In innovative psychoacoustic signal processing has ushered contrast, the results between 350 and 500 m3/h displayed in a transformative phase in the study of airflow dynamics more consistency without significant shifts. within forward-curved fans. This research paves the way for future endeavors aiming to optimize aerodynamic 4. CONCLUSIONS efficiency and noise control in various applications. Forward curved fans are widely utilized across various 4. REFERENCES industrial and domestic applications, and their operation has been thoroughly examined over the years. Despite [1] B. Eck, Fans, Oxford: Pergamon Press, 1973. this, the issue of efficiency reduction due to abrupt flow [2] B. Jiang, J. Wang, X. Yang and W. D. Y. Wang, “Tonal noise reduction by unevenly spaced blades in a forward-curved- conditions in the inlet channel remained largely blades centrifugal fan,” Applied Acoustics, vol. 146, pp. unexplored. This study primarily aimed to ascertain the 172-183, 2019. impact of inlet channel geometry on the fan's integral [3] C. Jian, H. Yuan, G. Li, W. Canxing, C. Liu and L. Yuanrui, aeroacoustic properties and to determine the nuances of “Aerodynamic noise prediction of a centrifugal fan local flow properties. This deeper understanding considering the volute effect using IBEM,” Applied subsequently provided insights into variations in integral Acoustics, vol. 132, pp. 182-190, 2018. [4] J. Gonzalez and e. all, “Symmetrized dot pattern analysis characteristics. The key insights and findings include: for the unsteady vibration state in a Sirocco fan unit,” 1. Efficiency and Noise Improvements: Through Applied Acoustics, vol. 152, pp. 1-12, 2019. modifications to the existing inlet channel, we [5] S. J. Kim, H. J. Sung, S. Wallin and A. V. Johansson, “Design achieved a 4% efficiency gain and a 2 dB(A) noise of the centrifugal fan of a belt-driven starter generator reduction. with reduced flow noise,” International Journal of Heat 2. Visualization Techniques: The integration of smoke and Fluid Flow, vol. 76, pp. 72-84, 2019. [6] Kind and R.J., “Prediction of flow behavior and visualization coupled with laser sheet illumination performance of squirrel-cage centrifugal fans operating at effectively revealed three distinct flow regions: a well- medium and high flow rates,” Journal of Fluids filled area with minimal vorticity, a turbulent eddy Engineering, vol. 119, 1997. region with limited throughflow, and an intermediate [7] R. Kind and M. Tobin, “Flow in a centrifugal fan of the zone with a pronounced continuous vortex. squirrel-cage type,” Journal of Fluids Engineering, vol. 112, 3. Hot Wire Anemometry: Traversing the inlet plane with 1990. [8] T. S. N. Adachi and e. All, “Study on the performance of a a hot wire anemometer provided essential data on sirocco fan (optimal blade design),” International Journal velocity fluctuations. When processed, this data of Rotating Machinery, vol. 7(6), 2001. significantly elucidated the flow patterns in relation to [9] T. Adachi, N. Sugita and e. all, “Study on the performance channel geometry and flow rate. of a sirocco fan (flow around the runner blade),” Journal Fluid Thermal Science, p. 13(3), 2003. 114 AAAA – 2023 – IZOLA - Conference Proceedings [10] R. Ballesteros-Tajaduraet and e. all, “Numerical model for [27] C. Bogey and C. Bailly, “An analysis of the correlations the unsteady flow features of a squirrel cage fan,” in Summer meeting, 2009. between the turbulent flow andthe sound pressure fields [11] S. Velarde-Suarez and e. all, “Experimental study on the of subsonic jets.,” Journal of Fluid Mechanics, vol. 583, pp. aeroacoustic behavior of a forward-curved blades 71-97, 2007. centrifugal fan,” Journal of Fluids Engineering, vol. 121, [28] J. Charonko, V. Cameron, L. Barton and P. Vlachos, 1999. “Assessment of pressure field calculations from particle [12] N. Montazerin, G. Akbari and M. Mahmoodi, image velocimetry measurements,” MEASUREMENT Developments in turbomachinery flow: Forward curved SCIENCE AND TECHNOLOGY, vol. 21, no. centrifugal fans, Woodhead Publishing in Mechanical Engineering, 2015. doi:10.1088/0957-0233/21/10/105401, pp. 1-16, 2010. [13] S. Pham Ngoc and e. all, “effects of bell mouth geometries [29] S. Obi and N. Tokai, “The pressure–velocity correlation in on the flow rate of centrifugal blowers.,” Journal of oscillatory turbulent flow between a pair of bluff bodies,” Mechanical Science and Technology, vol. 25(9), 2011. International Journal of Heat and Fluid Flow, vol. 27, pp. [14] M. Gholamian and e. all, “Effect of axial gap between 768-776, 2006. diffuser inlet and impeller on efficiency and flow pattern of [30] J. Pearse and M. Kingam, “Measurements of sound in centrifugal fans,” Applied Mechanics and Materials, pp. 446-447, 2013. airflow,” in The 13th International Congress on Sound and [15] Xuanfeng Wen, Datong Qi, Yijun Mao, Yijun Mao, Xin Yang. Vibrations, Vienna, Austria, 2. - 6. July 2006. Experimental and numerical study on the inlet nozzle of a [31] Wan, H.J. and Duck, J.L. A numerical study on the flow and small squirrel-cage fan. June 2013. Proceedings of the sound fields of centrifugal impeller located near a wedge, Institution of Mechanical Engineers Part A Journal of Power Journal of Sound and Vibration 266 (2003) 785–804 and Energy 227(4):450-463. DOI: 10.1177/0957650913487530 [32] Weidemann, J. Analysis of the Relation between Acoustic [16] and Aerodynamic Parameters for a Series of Dimensionally G. Lemaitre, C. Vartanian, C. Lambourg and P. Boussard, “A psychoacoustical study of wind buffeting noise,” Applied Similar Centrifugal Fan Rotors, NASA TT F–13, 1971, p. 798. Acoustics, vol. 95, pp. 1-12, 2015. [33] Wu, J.D. and Bai, M. R. Application of Feedforward [17] R. Penna-Leite, S. Paul and S. Gerges, “A sound quality- Adaptive Active Noise Control for Reducing Blade Passing based investigation of the HVAC system noise of an Noise in Centrifugal Fans, Journal of Sound and Vibration, automobile model,” Applied Acoustics, vol. 70, pp. 2001, 239 (5), 1051-1062, 636-645, 2009. [18] [34] Hyosung, S., Hyungki, S., Soogab, L. Analysis and Y. Okutsu and N. Hamamoto, “Effects of Hydrodynamic and Acoustic Pressure Fluctuations on Transmitted Sound Optimization of Aerodynamic Noise in a Centrifugal in Wavenumber-Frequency Domain,” in Internoise,, Compressor, Journal of Sound and Vibration, 2006 289, Melbourne, 16-19 November 2014. 999–1018 [19] P. Sagaut and C. Cambon, “Compressible Homogeneous [35] Čudina, M. and Prezelj, J. Noise Generation by Vacuum Isotropic Turbulence,” in Homogeneous Turbulence Cleaner Suction Units: Part I. Noise Generating Dynamics, doi:10.1007/978-3-319-73162-9_13 , Springer Mechanisms – An Overview, Applied Acoustics, 2007, 68 International Publishing, 2018, pp. 621-689. [20] (5), 491-502 H. Dryden and T. Karman, Advances in applied mechanics, New York: Academic press, 1964. [36] Wan, H.J. A Numerical Study on the Effects of Design [21] T. Suzuki and T. Colonius, “Instability waves in a subsonic Parameters on the Performance and Noise of a Centrifugal round jet detectedusing a near-field phased microphone Fan Journal of Sound and Vibration 2003, 265, 221–230, array,” J. Fluid Mech, vol. 565, pp. 197-226, 2006. [22] M. Mancinelli, T. Pagliaroli, R. Camussi and T. Castelain, J. [37] Steffen Marburg, Bodo Nolte, Computational acoustics of Fluid Mech., vol. 836, pp. 998-1008, 2018. [23] noise propagation in fluids – finite and boundary element S. G. M. F. Mario Felli, “A novel approach for the isolation of the sound and pseudo-sound contributions from near- methods, Springer Verlag, March 2008 field pressure fluctuation measurements: analysis of the [38] Daniel R. Raichel, The Science and Applications of hydroacoustic and hydrodynamic perturbation in a Acoustics, 14. Machinery Noise Control, pp. 357-408, propeller-rudder system,” Exp. Fluids, vol. 55, p. 1651, Springer New York, 2006, ISBN: 978-0-387-26062-4 2014. [39] N. Curle, The influence of solid boundaries upon [24] J. Ffowcs-Williams, “Noise source mechanisms,” in aerodynamic sound, Proc. Roy. Soc. London, A 231, pp. Modern Methods in Analytical Acoustics Lecture Notes, 505-514, 1955, Berlin, Springer, 1992, p. 313–354. [25] F. Auteri and e. all, “A novel approach for reconstructing [40] Fridolin P. Mechel, Formulas of acoustics, 2nd Edition, pressure from PIV velocity,” Exp Fluids, vol. 45, pp. 56-45, Springer Verlag, 2008 2015. [41] J. Prezelj et.all, Quantification of aerodynamically induced [26] O. Shinnosuke and T. Norihiko, “The pressure–velocity noise and vibration-induced noise ina suction unit. correlation in oscillatory turbulentflow between a pair of Proceedings of the Institution of Mechanical Engineers. bluff bodies,” Heat and fluid flow, vol. 27, pp. 768-776, 2006. Part C, Journal of mechanical engineering science, 2011, vol. 225, no. 3, pp. 617-624 115 Keynote Invited speech Development of sound quality metrics using models based on human perception and their applications Prof. Dr.-Ing. Roland Sottek Chalmers Applied Acoustics, Department of Architecture and Civil Engineering E-mail: roland.sottek@chalmers.se Professor Dr.-Ing. Roland Sottek has a position as Adjunct Professor in Psychoacoustics at the Division of Applied Acoustics at Chalmers University of Technology since 2016. He received a diploma in Electrical Engineering / Communications Engineering from the Technical University of Aachen in 1987 and a doctor’s degree in 1993 for his doctoral research study “Signal Processing Model of the Human Auditory System”. From 1987 to 1988 he worked as a scientist at the Philips Research Laboratory Aachen. In 1989 he joined HEAD acoustics where he was first Principal Scientist, later Head of the HEAD Consult NVH department and since 2002 Head of the newly established HEAD Research NVH department. In June 2023 he received the new role of Chief Scientific Advisor, directly supporting the Managing Director with scientific expertise. During his work at HEAD acoustics, he was involved in numerous consulting projects mainly related to automotive applications, as well as in 18 publicly funded national and international research projects. He is author or co-author of more than 150 publications and supervisor of more than 30 theses. Current research work concerns models of human hearing, psychoacoustics, localization and characterization of sound sources, auralization of virtual environments, noise engineering and digital signal processing as well as experimental and numerical methods for sound-field calculation. Sound quality metrics are often used to analyze complex sound scenarios, such as soundscape applications. Sound quality can also affect the health and well-being of people in a given environment. Therefore, it is of the utmost importance that the definition of good sound quality in a particular context is as precise as possible. In this aspect, psychoacoustic indicators are usually used to develop these metrics. In his lecture, Roland Sottek will review the existing standardized time-varying loudness models: the Zwicker method (ISO 532-1) and the Moore-Glasberg-Schlittenlacher method (ISO 532-3), which he supported as project leader and as ISO working group member, respectively, as well as the Sottek Hearing Model Loudness (recently standardized in ECMA 418-2 2nd Edition), a new approach based on a nonlinear combination of partial tonal and noise loudness (introduced in ECMA-74 17th Edition as part of the Sottek Hearing Model Tonality, now moved to ECMA 418-2) to better account for the fact that the loudness of tonal components, i.e., tonal loudness, may have a stronger impact on the loudness perception than the loudness caused by the noise components, i.e., noise loudness. Additionally, he will give an overview of the psychoacoustic roughness based on the Sottek Hearing Model (standardized in ECMA-418-2 1st and 2nd Edition) for the evaluation of 116 fast modulated sounds and on another model for slow modulated sounds: the Sottek Hearing Model Fluctuation Strength, which will be standardized in the near future. The talk will also provide insights into the complex mechanisms of forming overall noise assessments for some application examples with highly time-varying signals. 117 Contributed papers Advanced measurement techniques in acoustics 1. Challenges in the introduction of timbre coordinates for violoncelli Daniel Svenšek (University of Ljubljana, Faculty of Mathematics and Physics) 2. Soundscape monitoring system for earthquake-affected urban spaces – Zagreb case study Karlo Filipan (Catholic University of Croatia) 3. Immission Directivity as a tool for generation of Noise Maps Jurij Prezelj (University of Ljubljana, Faculty of Mechanical Engineering) 4. Data Selection for Reduced Training Effort in Vandalism Sound Event Detection Stefan Grebien (Joanneum Research) 5. Experimental sound field characterization with automated high-resolution impulse response measurements Rok Prislan (InnoRenew CoE) 118 CHALLENGES IN THE INTRODUCTION OF TIMBRE COORDINATES FOR VIOLONCELLI Daniel Svenšek1,2, Urša Kržič3, Rok Prislan4 1 Department of Physics, Faculty of Mathematics and Physics, University of Ljubljana, Slovenia 2 Laboratory for Molecular Modeling, National Institute of Chemistry, Slovenia 3 Music School Vrhnika, Vrhnika, Slovenia 4 Acoustic Laboratory, InnoRenew CoE, Izola, Slovenia Abstract: of even the smallest differences between instruments required Musical instruments are known for their subtle nuances of for this task. sound, and as bodies of considerable size, they have complex It is fair to admit that a scientific method, particularly in radiation patterns. It is therefore a challenge to capture their fields involving humanistic aspects like the one at hand, is inher-sound in a robust and reproducible manner while preserving the ently limited. It cannot substitute for the actual act of listening finest structures required for timbre analysis. In fact, timbre has to and even more so, physically testing an instrument. Neverthe-never been successfully used in practice as a measurable pa- less, it can be a valuable initial step and a preliminary reference rameter of an instrument, although the ability of humans, espe-point when it comes to personal preferences, rather than un ul- cially musicians, to perceive subtle differences in sound colour timate objective truth. Consider the significance that assigning between instruments of the same type has never been ques- concrete numerical values to describe the relevant sound prop- tioned. In recording cellos, we have succeeded in capturing their erties of musical instruments might have for their production spectra in such a way that we can introduce harmonic timbre and the market as a whole. The harmonic timbre coordinates coordinates. These are quantifiers that represent the harmonic we introduced have the potential to serve precisely as such nu-spectral aspect of an instrument’s sound in a musically relevant merical indicators, though certainly only partial. way. The basic challenges of introducing timbre coordinates are The pilot study is conducted with violoncelli. We stay almost presented using the cello as a case study, along with data pro-entirely away from psychoacoustics [1, 2], apart from the basic cessing steps required to generate timbre coordinates. The study connection between harmonic timbre and harmonic content of is important because the introduced coordinates have the potenthe power spectrum. Moreover, we adopt a front-end approach, tial to change the world of musical instruments by providing an concentrating solely on the instrument’s audible output during objective label for the harmonic sound color of each instrument. normal play, which includes the complete sound radiation pro- Keywords: musical instruments, harmonic timbre coordinates, cess. We intentionally avoid detailed mechanical analysis like linear timbre vector space the examination of sound or vibration modes or complex non- linear models of tone generation etc., addressed in numerous rigorous physical treatments [3–8] on the one hand and arising 1. INTRODUCTION in the practice of instrument making [6, 9] on the other. The differences between the sound of musical instruments of the same type, although certainly perceivable to sensitive ears, are extremely small. Small compared to the sensitivity of record- ing an instrument, related for example to the position of the 2. CHALLENGES OF THE WAVE FIELD microphone, the room and its acoustics, the instrumentalist’s playing etc. Our quest began with the question of whether a ro- The wave nature of the sound field entails hard physical facts bust, reproducible, and meaningful quantification of the timbre that cannot be avoided. On the one hand, we are dealing with of identical instruments is possible, in an intentionally reduced direct sound, i.e. with sound radiation with its complex radia- framework as would be required. Our goal now is to show that tion patterns [10–15]. On the other hand, we encounter the laws this is indeed possible, at least for harmonic timbre. A particular of the diffusive wave field [16–18], where care must and can be and central challenge is to obtain a universal, robust, yet sensi- taken to ensure that the recording space is large enough to ex- tive recording of a musical instrument that allows the analysis clude modal effects. Svenšek et al.: CHALLENGES IN THE INTRODUCTION OF TIMBRE COORDINATES FOR VIOLONCELLI 119 azimuthal angles, clm are general coefficients, and h(1) = j l l + inl (3) are the spherical Hankel functions of the first kind (describ- ing outgoing waves); jl and nl are the spherical Bessel func- tions of the first and second kinds (also known as the spheri- cal Bessel and Neumann functions). Eq. (2) represents the mul- tipole expansion of the pressure field of outward propagating waves, where l is the order of the multipole (monopole, dipole, quadrupole, octupole ... for l = 0, 1, 2, 3 ...) Assuming the radiating surface is a sphere (we pursue the general picture, not the details) with radius a, the radial compo- nent of the acoustic velocity v, which satisfies the equation of motion −iωρv = −∇p, (4) must match the radial velocity of the vibrating sphere at r = a. Let vr(θ, ϕ) = vlmYlm(θ, ϕ) be a term in the expansion of the latter over spherical harmonics (the value of the coefficient vlm depends on the angular shape of the oscillation). The boundary condition for the pressure is then Fig. 1. The typical recording configuration for violoncello ∂h(1)(kr) c l lm = iωρvlm (5) measurements. A microphone array of 33 microphones ∂r r=a designed and built for this purpose was used. The posi- and determines the coefficient c tioning of the musician and the microphones was main- lm of the solution Eq. (2). Let us determine these coefficients in the limit ka ≪ 1, i.e., tained between recordings. when the radius of the radiating sphere is small compared to the wavelength of the sound at a given frequency. In leading order, the Hankel function Eq. (3) is 2.1 Understanding the radiation field (2l − 1)!! We are interested in the physical picture of the radiation pro- h(1)(kr ≪ 1) ≈ in(1)(kr) ≈ −i , (6) l l (kr)l+1 cess, not the details of the radiation patterns, which vary greatly from one instrument to another. The key factor is the size of the where (2l − 1)!! = 1 × 3 × 5 × · · · × (2l − 1) and (−1)!! ≡ 1. radiating instrument, i.e., the characteristic size of the space en-It then follows from Eq. (5) closed by its radiating surfaces, relative to the considered wave- length of the radiated sound. The larger the instrument in com- (2l − 1)!! c lm i(l + 1)k = iωρvlm (7) parison to the wavelength, the greater the efficiency of radiation (kr)l+2 r=a and the more intricate the angular dependence of the radiation and finally pattern. cρv (ka)l+2 Let us inspect this in more detail. The radiated sound wave lm clm = . (8) (2l − 1)!! l + 1 p(r, t) = p(r)e−iωt with angular frequency ω and pressure am- plitude p is described like every sound/scalar wave by the am- In this small sound source limit, the amplitudes of all multipoles plitude equation decrease strongly with decreasing ka. Their ratios to the ampli- ∇2p + k2p = 0, (1) tude c00 of the monopole are where k = ω/c and c is the speed of sound. The general solu- clm vlm 1 (ka)l = , (9) tion is c00 v00 l!! l + 1 ∞ l again decreasing strongly with decreasing ka and for ka < 1 X X p(r) = c also with increasing l. lm h(1)(kr) Y l lm(θ, ϕ), (2) l=0 m=−l Note that the acoustic pressure field in the immediate vicin- ity of the sound source surface is that dictated by the velocity where Ylm(θ, ϕ) are spherical harmonics, θ and ϕ are polar and boundary condition via Eq. (4), i.e. with full angular dependence Svenšek et al.: CHALLENGES IN THE INTRODUCTION OF TIMBRE COORDINATES FOR VIOLONCELLI 120 of the source oscillation. When we move away, the contribu- for a cello, in the above example c4m/c00 becomes greater than tions of higher multipoles rapidly decrease due to the leading 0.1 for frequencies greater than about 275 Hz. dependence Eq. (6). However, if this decreasing dependence no When choosing the distance between the instrument and longer holds at the surface, there is also no longer a decline of the microphones, it is useful to make sure that the multipoles, the higher multipoles. which are strong only in the near field, decrease sufficiently. The In the vicinity of a small source (kr ≪ 1, near field), the near-field solution Eq. (10) shows that simply a/r must be suf- contribution of a multipole l to the total acoustic pressure Eq. (2) ficiently small. However, we cannot do more – the angular de-is thus pendences of the multipoles remaining in the far field cannot be eliminated and must be averaged by some strategy. X cρvlm (ka)l+2 . . . pl(r) = − Ylm(θ, ϕ) . . . l + 1 (kr)l+1 m 2.2 Understanding the diffuse field ka a l+1 = − ρcvlm Ylm(θ, ϕ). (10) l + 1 r First, we assume that the recording space is large enough to ex- clude modal effects – the choice of recording space is a control- At r = a it is as dictated by the oscillation of the source surface, lable factor. Similarly, the most prominent early reflections can and falls away from it as a power of a/r. Multipoles of higher be mitigated by a sufficiently large recording space and, in addi- order fall off faster with the distance from a small source. tion, by the use of absorbing materials on critical sections of the Far away from the source (kr ≫ 1, far field), the Hankel surfaces. What ideally remains, is the diffuse sound field. functions Eq. (3) take the asymptotic form A diffuse field is stochastically irregular and homogeneous on average, but one must be cautious about the true meaning h(1)(kr) ≍ i−(l+1) eikr (11) l kr of the average. At a given frequency, the diffuse field is by defini-and the contribution of a multipole l to the total acoustic pres- tion not homogeneous as it knows about a characteristic length sure Eq. (2) is – its wavelength. Indeed, it is spatially correlated on the wave- length scale, with spatial pressure correlation function [16,17,19] X pl(r) = clm i−(l+1) eikr Ylm(θ, ϕ). (12) kr ⟨p(r′, t) p(r′ + r, t)⟩ sin kr m χ(r) = = , (13) ⟨p2(r′, t)⟩ kr Note that the contributions of all multipoles, regardless of their order, fall off only as 1/r (as with any radiation field). This should where ⟨⟩ denotes averaging over space r′ and time t. not be confused with the rapid decay 1/rl+1 of static multipole That is, a monochromatic sound field is not actually diffuse fields. Thus, assuming that the coefficients vlm are not small, and requires spatial averaging, which is essentially statistical in the contributions of higher-order radiation multipoles are small nature. When such field is stationary (a standing wave), its am- only if their amplitudes clm are small – they do not become rel- plitude is a fixed and more or less regular function in space with atively weaker with distance! modulations on the wavelength scale, and there is little random- For a small sound source (ka ≪ 1), the coefficients Eq. (8) ness left. However, if the frequency or the configuration of the are small and so are their ratios Eq. (9) relative to the amplitude modal amplitudes change (e.g., due to a displacement or rota-of the monopole. Only in this case the l > 0 multipoles are tion of the sound source, reconfiguration of the room etc.), the negligible and sound radiation has no substantial angular depen- shape of the stationary field changes abruptly when the condi- dence, except in the immediate vicinity of the source. When the tions for a diffuse field are met. size of the sound source increases to ka ∼ 1 and beyond, this It is thus common to smooth the signal obtained from a is no longer true, as we already understand from the physical diffuse sound field by averaging its spectrum over frequency picture discussed after Eq. (9) and as also indicated by Eq. (9). bands. However, in the case of musical instruments, whose Note that Eqs. (6)-(10) were derived for ka ≪ 1. It turns out, tones consist of narrow and stable frequency peaks, this is not however, that they are nevertheless acceptable over a much possible – we cannot avoid the peculiarity of the superimposed larger range, e.g., up to ka ∼ 4 for l = 4 and ka ∼ 10 for monochromatic sound fields corresponding to harmonic par- l = 32, so that Eq. (9) serves as a reasonable estimate of the tials. Even in a large concert hall, where modal effects are com- relative strength of the multipoles. For example, c4m/c00 be- pletely absent, one can hear how individual harmonics of a dis- comes larger than 0.1 for ka >∼ 2. tant instrument playing a steady note become fainter and louder In general, for each radiation multipole l there is a critical when the listening position (or the position of the instrument) frequency above which that multipole is important and of undi- is slightly changed. Only spatial averaging can help here. minished relative amplitude even in the far field. Conversely, for each frequency there are significant multipoles with orders be-3. REQUIREMENTS FOR THE TIMBRE COORDINATES low a critical l. The angular dependence of these multipoles can- not be avoided. Therefore, the radiation pattern they form is in- If timbre coordinates are to be truly useful and have real value herently complex for sufficiently large ka. Assuming a ∼ 0.4 m in the musical sense, they must meet crucial requirements of Svenšek et al.: CHALLENGES IN THE INTRODUCTION OF TIMBRE COORDINATES FOR VIOLONCELLI 121 relevance, robustness, and reproducibility. i) They should correspond to the situation of normal playing by a musician. No artifi- cial excitation by devices is allowed. ii) It is essential that they remain robust and exhibit minimal variation when the same tone is played repeatedly in the same way. This inherent variation in C string cello 1 natural excitation defines a base tolerance, which represents the cello 2 upper limit of precision against which all other sources of errors cello 3 should be judged and which must be sufficiently smaller than cello 4 the timbre variations between instruments. iii) The variations cello 5 cello 6 arising from different performers should be minor in comparison 2 cello 7 to the variations between the instruments. Ideally, they should 1 cello 8 0.15 fall within the base tolerance margin, effectively rendering the 9 cello 9 0.10 cello 10 performer as an irrelevant factor. iv) The coordinates should be 7 6 11 0.05 cello 11 largely unaffected by natural variations that occur during normal e 3 5 3 4 cello 12 playing. These variations primarily stem from shifts in the posi- 0.00 10 timbr tion and orientation of the instrument during play, which influ- 0.05 ence the sound radiation pattern. v) The determination of the 12 0.10 coordinates must be reproducible, even in different measuring 0.15 spaces that comply with the standards. 8 0.15 0.1 0.10 0.0 0.05 0.1 4. MEASUREMENT AND ANALYSIS timbr 0.00 e 1 0.05 e 2 0.10 0.2 timbr Both the averaging of the sound radiation pattern and of the dif- 0.15 0.3 fuse field, discussed in Section 2, is achieved by recording with 0.2 mean timbre a sufficiently large number of microphones distributed across a characteristic listening solid angle in front of the instrument 0.1 and spanning a sufficiently large contour. A cross-shaped mi- 0.0 crophone stand was developed for the purpose, carrying 33 200 400 600 800 1000 1200 Hz equidistant microphones in a horizontal and vertical line. We 0.5 principal timbre 1 use phase-matched 1/4-inch microphones (B&K type 4958), de- 0.0 clared with a frequency response in the range of ±2 dB from 50 to 10 000 Hz. Data acquisition was performed with 24 bits 0.5 and a sampling frequency of 65 536 Hz (B&K Lan-XI data acqui- 200 400 600 800 1000 1200 Hz principal timbre 2 sition system). 0.5 The recording of the violoncelli was performed in a con- 0.0 trolled environment using the recording setup shown in the photo in Fig.1. The cellist bows a selected tone several times 0.5 in succession at a comfortable (mezzo) forte, where the respon- 200 400 600 800 1000 1200 Hz principal timbre 3 siveness of the instrument is best and the tone production is 0.5 most stable. The repetitions are used both for statistical averag- ing and to determine the base tolerance, which is the inherent 0.0 measure of uncertainty in the resulting harmonic timbre coordi- 0.5 nates. 200 400 600 800 1000 1200 Hz The analysis starts with the selection of the most stable seg- ments from the multichannel recordings of the sustained notes. Fig. 2. Top: The first three dimensions of the harmonic tim-For each channel, the power spectra of these segments are com- bre space of the violoncelli C strings. Bottom: harmonic puted with optional windowing. The powers within the har- vectors of the corresponding mean timbre and the princi- monic peaks are then integrated across their widths, averaged pal timbres – basis vectors of the above timbre subspace. across all channels, and optionally weighted to obtain multi- component vectors of the strengths of the harmonic partials, ty about 20. So many harmonics are usually not so stable and repro- ducible that they would lead to well-defined timbre coordinates. Svenšek et al.: CHALLENGES IN THE INTRODUCTION OF TIMBRE COORDINATES FOR VIOLONCELLI 122 Therefore, by stacking all harmonic vectors of the analyzed col- lection of instruments, properly normalized, into a matrix and performing the singular value decomposition (SVD) of this ma- trix (which we do not discuss in detail here), we can determine A string cello 1 a smaller number of relevant combinations of harmonics – the cello 2 so-called principal timbres [20,21]. They define a statistically op-cello 3 timum basis of the harmonic timbre vector space. The principal cello 4 timbres are not known in advance and depend on the type of 0.25 cello 5 instrument, the register, the string, the dynamics of the playing 0.20 cello 6 and other possible variations, which then define specific princi- timbr 0.15 cello 7 7 pal timbre bases and specific sets of coordinates. cello 8 e 3 0.10 cello 9 After “mining” these principal timbres solely from the front- 0.05 11 cello 10 end acoustic data, one can also listen to them and form his own 0.00 4 5 cello 11 psychoacoustic interpretations. Furthermore, one can envision 0.05 8 cello 12 an instrument in which the principal timbres are added with dif- 0.10 2 ferent weights (positive and negative). One can then explore in- 0.15 12 9 struments whose coordinates align with these specifications. 0.15 10 1 0.10 3 6 timbr 0.05 5. EXAMPLES OF RESULTS 0.00 e 2 We will present selected results based on a collection of 12 in- 0.05 0.10 struments recorded in the main concert hall of the Conservatory 0.15 of Music and Ballet Ljubljana in May 2023 (many thanks to the 0.25 0.20 0.15 0.10 0.05 0.00 0.05 0.10 Conservatory, Prof Karmen Pečar and her students). We ana- timbre 1 lyzed the timbres of the four open strings of the violoncelli, thus 0.4 mean timbre characterizing each instrument by four timbres, one for each string. Above in the Figs. 2 and 3 are shown the resulting har- 0.2 monic timbre spaces for the lowest (C) and highest (A) strings of 0.0 the cello, more precisely, their dominant three-dimensional sub- 1000 2000 3000 4000 Hz spaces. The individual instruments form well-defined clusters in 0.5 principal timbre 1 this subspace. The centre of the ellipsoid represents the mean 0.0 of 10 repeated strokes of the open string of the corresponding cello, while the ellipsoids indicate the statistical uncertainty in 0.5 each timbre direction, i.e., the base tolerance of playing. 1000 2000 3000 4000 Hz principal timbre 2 For some instruments the tolerances are very small, for oth- 0.5 ers much larger, especially for the A string in Fig. 3 (top). These are not only uncertainties of the determined timbre coordi-0.0 nates, but also suggest differences in the responsiveness (playa- 0.5 bility) of the instruments. Some of the ellipsoids are highly elon-1000 2000 3000 4000 Hz principal timbre 3 gated (prolate), usually indicating pronounced variations of a 0.5 particular single harmonic partial with each repetition of the tone. In principle, though less likely, it is also possible that two 0.0 or more partials vary that happen to lie in similar directions of 0.5 the represented subspace (but the directions of the individual 1000 2000 3000 4000 Hz partials are always orthogonal in the complete timbre space). The decisive importance of spatial averaging of the sound Fig. 3. The same diagrams as in Fig. 2, but for the A strings of an instrument is demonstrated in Fig. 4. Here, the harmonic of the celli. Note the considerable differences between the vectors of the individual channels are projected onto the tim- sizes and shapes of the uncertainty ellipsoids, discussed in bre subspaces of Figs. 2 and 3, respectively, without changing the text. the spaces themselves. It is obvious that the variations between the channels, even if they have a hint of regularity and are not completely stochastic, are extremely large and completely blur the well-defined clusters formed by the averages of the channels Svenšek et al.: CHALLENGES IN THE INTRODUCTION OF TIMBRE COORDINATES FOR VIOLONCELLI 123 C string – see below in Fig.4 what is left without the aid of supervised colour coding. Without spatial averaging there are no timbre coordinates! 2 0.15 1 6. CONCLUSION 9 0.10 We have highlighted some of the challenges associated with the 0.05 11 e 3 7 pioneering attempts to quantify minute differences in the har- 6 4 0.00 monic timbre of identical instruments. We presented some of 5 timbr 3 cello 1 10 0.05 the solutions to these challenges. For the measurement part cello 2 0.10 of the process, this is primarily the issue of adequate spatial cello 3 12 averaging of an instrument’s sound field. When properly aver- cello 4 0.15 cello 5 aged, the introduced harmonic timbre coordinates are robust – 8 0.1 cello 6 their determination is reproducible almost entirely within the in- cello 7 0.0 herent tolerance of human playing. Ongoing experiments have 0.15 0.10 cello 8 0.1 cello 9 0.05 e 2 shown that they are also largely independent of the measure- cello 10 timbr 0.00 0.2 timbr ment space, with a somewhat extended tolerance, provided the e 1 0.05 0.10 cello 11 0.15 0.3 space is not too small – a chamber hall already proves to be large cello 12 enough. Systematic studies to refine the definition of the toler- A string ances are underway. 7. REFERENCES 0.25 0.20 [1] S. Town and J. Bizley, “Neural and behavioral investigations timbr 0.15 into timbre perception,” Frontiers in Systems Neuroscience, 7 e 3 0.10 vol. 7, 2013. 0.05 11 [2] S. McAdams and K. Siedenburg, “Perception and cognition 0.00 5 4 of musical timbre,” in Foundations of Music Psychology: The- 0.05 8 ory and Research (P. J. Rentfrow and D. J. Levitin, eds.), 0.10 2 0.15 12 pp. 71–120, Cambridge: MIT Press, 2019. 9 0.15 [3] M. E. McIntyre and J. Woodhouse, “On the fundamentals of 10 0.10 3 1 bowed string dynamics,” Acustica, vol. 43, no. 2, pp. 93–108, timbr 0.05 6 1979. 0.00 e 2 [4] M. E. McIntyre, , R. T. Schumacher, and J. Woodhouse, “On 0.05 the oscillations of musical instruments,” J. Acoust. Soc. Am. , 0.10 vol. 74, no. 5, p. 1325, 1983. 0.15 0.25 0.20 0.15 0.10 0.05 0.00 0.05 0.10 [5] L. Cremer, The physics of the violin. Cambridge, Mass: MIT timbre 1 Press, 1984. A string [6] C. M. Hutchins and V. Benade, eds., Research Papers in Violin Acoustics, 1975-1993. Acoustical Society of America, 1996. [7] N. H. Fletcher and T. D. Rossing, The Physics of Musical In-0.25 0.20 struments, 2nd Edition. New York: Springer, 2010. 0.15 0.10 e 3 [8] A. Chaigne and J. Kergomard, Acoustics of musical instru- 0.05 timbr ments. Springer, 2016. 0.00 0.05 [9] CAS — Catgut Acoustical Society, 0.10 0.15 https://www.catgutacoustical.org. 0.15 0.25 0.10 [10] G. Weinreich and E. B. Arnold, “Method for measuring 0.20 0.05 0.150.10 0.00 0.05 e 2 acoustic radiation fields,” The Journal of the Acoustical So- timbr 0.05 e 1 0.00 0.10 timbr 0.05 0.10 0.15 ciety of America, vol. 68, no. 2, pp. 404–411, 1980. Fig. 4. Harmonic vectors of the individual microphone channels of the measuring array, projected onto the cor- responding timbre subspaces of Figs. 2 and 3. Bottom: the A string diagram as above, but without the colour coding. Svenšek et al.: CHALLENGES IN THE INTRODUCTION OF TIMBRE COORDINATES FOR VIOLONCELLI 124 [11] L. M. Wang and C. B. Burroughs, “Directivity patterns of in reverberant sound fields,” The Journal of the Acoustical acoustic radiation from bowed violins,” Catgut Acoust. Soc. Society of America, vol. 27, no. 6, pp. 1072–1077, 1955. J. , vol. 3, no. 7, pp. 7–15, 1999. [17] F. Jacobsen, The diffuse sound field - Report No. 27. The [12] G. Bissinger, “Parametric plate-bridge dynamic filter model Acoustic Laboratory, Technical University of Denmark, 1979. of violin radiativity,” The Journal of the Acoustical Society of [18] B. Rafaely, “Spatial-temporal correlation of a diffuse sound America, vol. 132, no. 1, pp. 465–476, 2012. field,” The Journal of the Acoustical Society of America, [13] N. R. Shabtai, G. Behler, M. Vorländer, and S. Weinzierl, vol. 107, no. 6, pp. 3254–3258, 2000. “Generation and analysis of an acoustic radiation pattern database for forty-one musical instruments,” The Journal of [19] F. Jacobsen and T. Roisin, “The coherence of reverberant the Acoustical Society of America, vol. 141, no. 2, pp. 1246– sound fields,” The Journal of the Acoustical Society of Amer- 1256, 2017. ica, vol. 108, no. 1, pp. 204–210, 2000. [14] J. Meyer, Acoustics and the Performance of Music. Springer [20] G. J. Sandell and W. L. Martens, “Perceptual evaluation of New York, NY, 2009. principal-component-based synthesis of musical timbres,” J. [15] J. Pätynen and T. Lokki, “Directivities of symphony orches- Audio Eng. Soc. , vol. 43, pp. 1013–1028, december 1995. tra instruments,” Acta Acustica united with Acustica, vol. 96, [21] M. A. Loureiro, H. B. Paula, and H. C. Yehia, “Timbre classifi-no. 1, pp. 138–167, 2010. cation of a single musical instrument,” in In Proc. Intl. Conf. [16] R. K. Cook, R. V. Waterhouse, R. D. Berendt, S. Edelman, and on Music Information Retrieval (ISMIR, 2004. M. C. Thompson, “Measurement of correlation coefficients Svenšek et al.: CHALLENGES IN THE INTRODUCTION OF TIMBRE COORDINATES FOR VIOLONCELLI 125 SOUNDSCAPE MONITORING SYSTEM FOR EARTHQUAKE-AFFECTED URBAN SPACES – ZAGREB CASE STUDY Karlo Filipan1, Mia ˇ Setić Beg2, Dominik-Borna Ćepulić2, Hrvoje ˇ Stefančić2 1 Acoustics Lab, Catholic University of Croatia, Zagreb, Croatia 2 Department of Psychology, Catholic University of Croatia, Zagreb, Croatia Abstract: tention to and assign the meaning to, affect their experience of The experience of a sound stimulus is closely related to the mean-the place [6]. ing that people attach to it. In Zagreb area, after the series of In 2020, Croatia was hit by two major earthquakes in the city earthquakes which struck in 2020, people have started reporting of Zagreb and Banovina county. In addition to material damage, the increased noticing and reaction to the sounds similar to the people have also experienced increased stress and anxiety levels ones produced by earthquakes. During the last year, project In-induced by these earthquakes [7]. The earthquake events were SPE(S) was started with the goals: a) to establish the infrastruc-felt by the strong vibrations of the built environment as well as ture for measuring sound and vibrations in earthquake-affected by the sound that accompanied the earthquake [8]. Due to this urban areas, and b) to examine the connection between objec- traumatic experience, people started reacting even to the lower tive characteristics of sound and vibrations with people’s percep-level stimuli such as the passing of a heavy vehicle, the slamming tion, personality traits and previous experiences of earthquakes. of doors and windows, the passing of an elevator, and the like. Monitoring methodology is structured in two ways: sound and In order to quantify such experiences, a long-term monitor- vibration measurements are performed using a network of sen- ing system for sound and vibration measurements and instanta- sor nodes, while respondents who live near the sensors provide neous soundscape appraisal is being established during project their experience of the salient events captured by the sensors In-SPE(S) - Investigation and Measurement of Soundscape through a mobile application. This contribution will present the Perception in Earthquake-affected Urban Spaces. The data ob-soundscape monitoring system and research methodology uti- tained by the system will be used to examine the connections lized in the project as well as discuss some preliminary results of between physical characteristics of sound and vibrations with the measurements campaign. people’s perception, personality traits and previous experiences Keywords: soundscape, sensor network, mobile application, of earthquakes. In this contribution, we will give an overview of physical measurements, perception evaluation the established soundscape monitoring system as well as discuss some preliminary results and ongoing work. 1. INTRODUCTION 2. SOUND AND VIBRATION MEASUREMENTS It is well-understood that people living in highly urbanized areas mention noise as one of the biggest nuisance factors of their In the proposed soundscape monitoring system (Fig. 1) sound lives. Since the second half of the 20th century, the negative ef- and vibrations are measured by a network of sensor nodes fects of noise on health and human behavior have been system- placed on the residential buildings. The sensor node (Fig. 2) atically investigated [1]. However, in addition to the research on consists of a Raspberry Pi 3 Model B+ single-board computer noise, the concept of soundscape has also been developed dur- to which various sensors and electronic components are con- ing the last 50 years [2]. In the more recent ISO standard [3], nected to. The sensors are: integrated chip designed by the soundscape is defined as the: “acoustic environment as per- ASAsense company [9] with Knowles FG-23329-P07 microphone ceived or experienced and/or understood by a person or peo- and ADXL345 accelerometer; and a DHT22 sensor used for mea- ple, in context”. Thus, soundscape analyzes detailed interac- surements of temperature and humidity. All electronics are se- tion of sound, environment and people, including both positive cured in a waterproof housing. The node is connected to elec- and negative effects [4]. Research has shown that the percep- tricity via power supply cable and to the Internet via WiFi. To tion of the sound environment and consequent soundscape ap- obtain comparable sound measurements, a class 1 acoustic cal- praisal depends on the experience of prominent (salient) sound ibrator SVANTEK SV 36 is used to calibrate the microphone. sources [5]. Therefore, the sounds that people hear, i.e. pay at- The recorded sound and vibration measurements are stored Filipan et al.: Soundscape monitoring system for earthquake-affected urban spaces – Zagreb case study 126 Fig. 1. Soundscape monitoring system established during Project In-SPE(S): physical environment is measured by the sensor nodes and data is analyzed on the server (red); peo- ple are asked for their appraisal of salient events through mobile application (blue) and their answers are stored for further analysis and use (green). using a time-stamped data structure in the memory of the sen- sor node. The stored data is compressed and encrypted. This en- sures that, in accordance with the GDPR directive, the data sent to the cloud cannot be read by third parties. Rclone software tool [10] is responsible for continuously encrypting and sending the measurements to the cloud as well as storing them into a directory structure according to the sensor’s tag (location) and the time of recording. Fig. 2. Top: placement of sensor nodes on external build-Measurements from the sensors are analyzed on the Linux ing envelope. Bottom: operational setup of the sensors server. The rclone tool is again used to download and decrypt (microphone, temperature and humidity sensor, sound the data from the cloud. Some examples of the parameters and vibration acquisition board) connected to a single- continuously measured by the system can be seen in Figs. 3-6. board computer (Raspberry Pi 3 Model B+). Firstly, temperature and relative humidity are parameters which need to be accounted for when making the statistical models that include sound and vibration [11]. As it can be seen in Figs. 3 and 4, placement of the sensor also plays a role on their variabil-on the building close by but somewhat shielded from the ma- ity. Hence, although somewhat influenced by the enclosure and jor arterial road, while node 2 is placed on the building looking the heat from the electronics, measuring these values is use- over a side road. Hence, daily LAeq levels are higher for node ful. Moreover, they could also indicate potential faults of the 2. On the other hand, levels during night period, 23h-7h, drop equipment. For example, for sensor 1 (Fig. 3, bottom), humidity more in the case of node 2. Regarding the frequency content, raised up significantly at 15 o’clock on a particular date. Knowing LCeq − LAeq levels measured by node 1 (Fig. 5, bottom), show the local situation (rain shower), the leak of the enclosure was rush hour peaks (visible only during weekdays) which are not detected and accounted for. present in the side road as measured by node 2 (Fig. 6, bottom). From the sound measurements, indicators such as L Moreover, low frequency content is present largely during night Aeq and L periods for both measurement locations, although for the arte- Ceq can be calculated. A-weighted equivalent continuous level represents the overall sound/noise level, while the difference rial road this is not as prominent as for the side road. between the C-weighted and A-weighted levels indicates the This brief analysis shows that by using the sensor measure- low-frequency content typically associated with the traffic noise. ments and simple statistical indicators, sound environments can As shown in Figs. 5 and 6, there is a difference between the be differentiated and explained up to a point. Nevertheless, overall sound levels for sensor nodes 1 and 2. Node 1 is placed people living next to the sensor node placed by the arterial road Filipan et al.: Soundscape monitoring system for earthquake-affected urban spaces – Zagreb case study 127 Fig. 5. Daily variability of continuous sound levels mea-Fig. 3. Daily variability of environmental parameters measured by the sensor node 1 - close by a major arterial road: sured by the sensor node 1 - placed on the window facing top) A-weighted levels; bottom) difference between C-east: top) temperature; bottom) relative humidity. weighted and A-weighted levels. (Night times are shaded, and weekends and holidays are plotted with dashed lines.) Fig. 6. Daily variability of continuous sound levels mea-Fig. 4. Daily variability of environmental parameters measured by the sensor node 2 - building in a side road: top) Asured by the sensor node 2 - placed on the window facing weighted levels; bottom) difference between C-weighted west: top) temperature; bottom) relative humidity. and A-weighted levels. (Night times are shaded, and weekends and holidays are plotted with dashed lines.) Filipan et al.: Soundscape monitoring system for earthquake-affected urban spaces – Zagreb case study 128 (Fig. 5), report disturbance from night pass-bys of fast and loud motorcycles and cars. What is more, the passing trucks also of- ten produce tire whistling noise and, due to the exit from the underpass, loud rattling noise and vibrations coming from their bulk cargo. Hence, in order to better evaluate the soundscape of the place and to relate it to people’s perception, more detailed mod- els need to be used. Thus, soundscape monitoring system was established for long-term monitoring which includes continuous recording of sound and vibration signals. Hence, in addition to the sensor nodes that gather the data, the server infrastructure was created to continuously acquire and evaluate the obtained measurements. Therefore, for the purpose of extracting infor- mation from the data, already published and available statisti- cal, computational and machine learning models (e.g. [12–14]) for detection and labeling of the salient, i.e. noticeable events will be evaluated in the next phase of the project. 3. MEASURING PERCEPTION USING MOBILE APPLICATION Another part of the soundscape monitoring system is the mobile application implemented for the Android operating system (Fig. Fig. 7. Screenshots of the Android app: left) initial screen 7). The mobile app is used to acquire the in-situ and (almost) for starting up the questionnaire; right) question on self- immediate reactions to the sound and vibration events. The use reported anxiety level. of the app is twofold: the participants can label the sound and vibration events which they notice as well as assess the events which are measured by the sensor nodes. The application will be used by the participants taking part Participants will be notified that, at the end of the project, the in the case study. Each participant will give their informed con-assigned budget will be split proportionally between the asso- sent prior to using the application. Participants will also be given ciations based on the points collected by the participants. An-unique registration keys for logging into the application. To pro- other gamification characteristic of the app is the timer that tect their privacy, the keys will not be linked to their personal counts down the time until the questionnaire for a particular information but rather generated randomly and handled by the event is active. This motivates the participant to answer and research team. gather points relatively quickly and ensures that the information The app continuously monitors the signal from the server given is related to the event in question. which sends the location and the time of the sound and vibration events detected by the sensors. In case the mobile phone (i.e. the participant) is inside the radius of the sensor node on which an event was detected, the application shows that something 4. CONCLUSIONS had happened (Fig. 7, left). The respondent is then taken to a short series of questions (Fig. 7, right) asking about their: notic-In this contribution, we have presented an overview of the ing of the event, event characterization (intensity, pleasantness soundscape monitoring system that is being established and and source category) and evoked emotions (anxiety and distur- used in project In-SPE(S). The system will enable long term mon- bance from the event). The results of each questionnaire are itoring and evaluation of soundscape in Zagreb area which was then sent to a protected local database. Using the same proce- affected by the earthquakes. dure, the participants have the opportunity to initiate the ques- Currently, the work on the statistical models for prediction tionnaire themselves thus reporting the event they had noticed. of noticeable events is ongoing. The measurements and testing To motivate the participants to provide their answers, the of the system are performed on the campus of the Catholic Uni- mobile application has been implemented with the gamification versity of Croatia. In the short term, the participants who will and social involvement processes in mind. Specifically, during use the mobile application will also be recruited from the stu- the login process each user selects a non-profit association to dent population. This would provide a somewhat controllable which their answers are assigned to. Each time an answer is setting and enable tuning of the models as well as the whole stored, a number of points is given to the selected association. soundscape monitoring system. Filipan et al.: Soundscape monitoring system for earthquake-affected urban spaces – Zagreb case study 129 5. ACKNOWLEDGEMENT ing Events and Well-Being in Croatia, Psychological Reports, 00332941231156813, 2023. This study was funded and supported by an approved research project of the Catholic University of Croatia: “Investigation and [8] Tosi, P., Sbarra, P., and De Rubeis, V. Earthquake sound per-Measurement of Soundscape Perception in Earthquake-affected ception, Geophysical research letters, 39(24), 2012. Urban Spaces”. [9] Van Hauwermeiren, W., David, J., Dekoninck, L., Pessemier, T. D., Joseph, W., Botteldooren, D., Martens, L., Filipan, K., 6. REFERENCES and Coensel, B. D. Assessing road pavement quality based on opportunistic in-car sound and vibration monitoring, In [1] Azrin, N. H. Some effects of noise on human behavior, Jour-26th International Congress on Sound and Vibration (ICSV nal of the Experimental Analysis of Behavior, 1(2), 183, 1958. 2019), 2019. [2] Schafer, R. M. The soundscape: Our sonic environment and [10] Breitinger, F., Zhang, X., and Quick, D. A forensic analysis of the tuning of the world. Simon and Schuster, 1993. rclone and rclone’s prospects for digital forensic investiga- [3] ISO 12913-1: 2014. Acoustics – Soundscape – Part 1: defini- tions of cloud storage, Forensic Science International: Digi-tion and conceptual framework, International Organization tal Investigation, 43, 301443, 2022. for Standardization, 2014. [11] Van Hauwermeiren, W., Filipan, K., Botteldooren, D., and [4] Aletta, F., Kang, J., and Axelsson, Ö. Soundscape descrip- De Coensel, B. A scalable, self-supervised calibration and tors and a conceptual framework for developing predictive confounder removal model for opportunistic monitoring of soundscape models, Landscape and Urban Planning, 149, road degradation, Computer-Aided Civil and Infrastructure 65-74, 2016. Engineering, 37(13), 1703-1720, 2022. [5] Kang, J., Aletta, F., Gjestland, T. T., Brown, L. A., Bottel- [12] ISO 532-1: 2017. Acoustics – Methods for calculating loud- dooren, D., Schulte-Fortkamp, B., Lercher, P., van Kamp, I., ness – Part 1: Zwicker method, International Organization Genuit, K., Fiebig, A., Coelho, J. L. B., Maffei, L., and Lavia, for Standardization, 2017. L. Ten questions on the soundscapes of the built environ- [13] Sottek, R., and Lobato, T. H. G. Tonal Component Separation ment, Building and environment, 108, 284-294, 2016. of e-Vehicles Using the High-Resolution Spectral Analysis [6] Filipan, K., Boes, M., De Coensel, B., Lavandier, C., Delaitre, (HSA), SAE Technical Paper No. 2023-01-1141, 2023. P., Domitrović, H., and Botteldooren, D. The personal view- [14] Bibbo, G., Singh, A., and Plumbley, M. D. Audio Tag- point on the meaning of tranquility affects the appraisal of ging on an Embedded Hardware Platform, arXiv preprint the urban park soundscape, Applied Sciences, 7(1), 91. arXiv:2306.09106, 2023. [7] Prizmić-Larsen, Z., Vujčić, M. T., and Lipovčan, L. K. Fear of COVID-19 and Fear of Earthquake: Multiple Distress-Filipan et al.: Soundscape monitoring system for earthquake-affected urban spaces – Zagreb case study 130 Immission Directivity as a tool for generation of Noise Maps Jurij Prezelj, Jure Murovec, Anže Železnik, Luka Čurovič University of Ljubljana, Faculty of mechanical engineering, Aškerčeva 6, 1000 Ljubljana, Slovenia Abstract: This research unveils a pioneering method for estimating the sound power level of elements on an emission plane, utilizing a network of immission directivity sensors on a parallel receiving plane, yielding toward measurement-based noise maps. Confronting the challenges of the ill-posed inverse problem in acoustics, we introduce an intuitive algorithm that draws from the statistical analysis of environmental noise source behaviour in the time domain and statistics of their spatial distribution. Assuming the monopole character of environmental noise sources, we employ the basic equation for correlation between sound pressure and sound power above partially absorptive surface. By statistical analysis of the set of the calculated sound powers for each element on the emission plane, we regularize the problem, rendering it well-posed. This heuristic approach, reminiscent of optimization techniques, requires further validation against more extensive experimental dataset. Our methodology, grounded in acoustics and signal processing principles, enables development of real-time visualization of the spatial distribution and intensity of industrial noise sources. Remarkably, only a few sensors are necessary to compute a noise source map, or sound power map, which can subsequently be converted into a traditional sound pressure map. This innovative solution addresses a typically unresolved inverse problem in acoustics. Although our preliminary results are limited, they serve as a proof of concept, indicating the potential of this technique in enhancing environmental noise management, bolstering noise control measures, and guiding the design of future industrial sites to mitigate noise impacts. Keywords: Environmental noise, Noise mapping, Microphone differential array, Immission directivity 1. INTRODUCTION 1.1. Problem identification In Europe, traffic noise pollution, costing over €2.7 billion The current methodology for assessing the impact of annually, ranks as the second leading environmental noise on people primarily employs noise maps, which are cause of health problems [1-3]. To counteract this, the EU derived from official yearly averages of road, air, and rail adopted the "Towards Zero Pollution for Air, Water, and traffic data. This approach is designed to ascertain the Soil" Action Plan as part of the European Green Deal. A number of individuals subjected to high levels of noise core goal of this plan is a 30% reduction in the population annoyance, sleep disturbances, and associated health chronically disturbed by traffic noise [4]. The methodology risks. Regrettably, it overlooks the influence of other noise we propose addresses this issue by integrating noise sources like industrial noise from ports, wind power monitoring into noise mapping. Leveraging IoT plants, industrial premises, heat pumps, and air microphone arrays to measure "immission directivity", we conditioners [5,6,7]. Additionally, the diversity of noise can generate advanced noise maps, pinpoint noise types within large ports necessitates their classification sources, and effectively manage noise pollution. into five macro categories - road, railways, ship, port, and 131 AAAA – 2023 – IZOLA - Conference Proceedings industrial sources [5,6]. Despite their utility, noise maps focused on the development of low-cost sensors over the present several limitations. The precision of these maps in last fifteen years, ranging from proof-of-concept to the predicting noise levels in shielded or quiet areas is deployment of operational networks [8]. Several projects relatively low, and they often fail to encapsulate less based on low-cost environmental noise sensor networks predictable sound sources [8]. Further, the propagation have been developed in recent years [20,18,21,22,23,24]. models utilized in commercial software packages are For instance, Alvares-Sanches et al. [25] see an integrated grounded in a number of approximations, such as network of static sensors across a city to be the future, disregarding diffusion through facades and fitting objects, while Guillaumea et al. [7] merged noise modelling with and barely taking into account urban micro- measurements from fixed and mobile locations to extract meteorological conditions and vegetation [8]. Current the noise that is not generated by traffic. Benocci et al. methodologies do not consider the cumulative exposure [26,27,19] presented an automatic monitoring system to multiple noise sources [5,9] or interactions between based on customized low-cost sensors and a software tool distinct noise sources, such as the interplay between road implemented on a general-purpose GIS platform. This traffic and industrial noise, or between natural sound system performs the update of noise maps in real-time sources and road traffic noise [5,10]. Most significantly, (dynamic noise maps) on a large urban area by scaling pre- they do not account for fluctuations in noise levels calculated basic noise maps [27]. Balastegui et al. [28,29] [8,11,2,12,13,14], which can amplify the disturbing effects demonstrated that mobile sampling could be an option to of noise and play a crucial role in the evaluation of high increase the spatiotemporal coverage of samples to noise annoyance and sleep disturbances [9,11,12]. While produce a noise map with fewer resources. Furthermore, strategic noise maps are an effective tool for identifying mounting a noise sensor in a bicycle showed good critical areas and proposing mitigation plans, they fall performance for the case of roads with high traffic [28]. short when it comes to assessing spatiotemporal traffic Gajardo et al. [30] showed that a measurement duration conditions [15]. Thus, there is a compelling need for an of one week would provide sufficient data to estimate a innovative, robust methodology that can promptly yearly average and assess noise level fluctuations with an differentiate among various noise sources and provide overall error of less than 2 dB. Benocci et al. [26] showed insight into their spatiotemporal distribution [12]. A that measurements of SPL with low-cost measuring deeper understanding of the primary factors influencing microphones provide a correlation R2=0.9908 with urban noise will guide the formulation of more effective measurements of SPL with Class 1 Sound level meters, noise control policies [16]. The development of a allowing for a large number of low-cost sensors to be measurement system compatible with this novel used. Mahapatra et al. [31] showed that an absolute error methodology would furnish objective data on noise for source localization of less than 3m can be achieved exposure in urban environments, thereby enabling more with six microphones in the network if distances between accurate assessments of smaller areas of interest over microphones are 35 m, and the source is also 35 m from time. the network axis. As a general rule, the lower the cost of the sensor, the more sensors must be used, and the more 1.2 State of the art in recent studies spatially accurate the data will be [20]. Quintero et al. [29] Asdrubal et al. [17] provided a review of four confirmed that with low-cost hardware, it is possible to promising and emerging research areas for assessing generate reliable noise measurements with an equivalent sound environments in urban settings: dynamic noise accuracy to a Class 2 sound level meter for LAeq, and with maps, smart sensors, and smartphones, and the a Type A standard uncertainty very similar to the one soundscape approach. These areas are not yet fully observed by the reference equipment. When measuring matured from both a technological and legislative point of the sound pressure level (SPL) of a target noise source in view. However, recent technical developments such as an outdoor environment, a single microphone cannot the miniaturization of electronic components, efficiently select a target signal from a mixed sound accessibility of low-cost computing processors, improved source. Therefore, a microphone array can be used to performance of electric batteries, innovations in enhance the target signal through beamforming, communication networks, and acoustic signal processing effectively restraining the interference of non-target have opened new prospects for the deployment of low- signals [32]. Okatidiowo et al. [33] designed a cost sensor networks [8,18]. Low-cost sound recorders are beamforming antenna sensor for environmental noise also available to draw a real picture of the sound detection to discriminate vehicle emission according to environment over an extended area at a cost-effective road conditions. Leiba et al. [34] proposed a monitoring solution [19]. Consequently, many researchers have system of urban areas based on the use of large Prezelj et al.: Immission Directivity as a tool for generation of Noise Maps 132 AAAA – 2023 – IZOLA - Conference Proceedings microphone arrays to extract the radiated sound field sources propagates to a parallel immission plane. Each from each passing-by vehicle in typical urban scenes. They partial surface of this parallel plane can be regarded as an trained a machine learning algorithm to classify these immission point, 𝑝𝑟𝑚𝑠,𝑗. All the noise sources collectively extracted signals into clusters combining both the vehicle contribute to the sound level at each immission point, as type and driving conditions. This system makes it possible shown in Fig. 1. Let us assume the distance between the to monitor the evolution of the noise levels for each planes h to be relatively small in comparison to the cluster [34]. observation area dimensions and falls within the range of 0.5 a < h < 2 a, where ' a' is the dimension of the discrete A review of the literature indicates a pressing need for elements on both planes. For simplicity in this paper h, time-dependent noise mapping derived from and a will be set to 1 m. Such a simplification is feasible measurements. However, the sheer number of because environmental noise sources (traffic, industrial measurement locations required for this makes it sites, wind turbines, etc.) are typically distant from the impractical. Thus, we propose a novel concept for noise measurement point. As a result, the direction of mapping that leverages measurements of immission propagating waves can be considered parallel to the directivity which significantly reduces the number of ground. necessary measurement locations. 2. THEORETICAL BACKGROUND The methodology we present is rooted in the concept of measuring the Immission Directivity at different locations. This is accomplished by using a network of differential microphone arrays. The idea of using a differential microphone array comes from its capacity to differentiate the direction of incoming sound, by measuring the arrival times of sound to microphones in Fig. 1: Emission and Immission plane the array. Thus, the primary function of individual array in the network is to pinpoint the direction of the temporal Environmental noise sources are uncorrelated, and hence, noise source and to quantify its intensity. the sound waves they generate are incoherent. The probability density function of a single monopole source The proposed methodology, which is based on the is random and unique. It can be described in terms of the network of differential microphone arrays, enables probability of its occurrence, including the conditions for gathering measurements in real-time, which allows the its non-operation, [44]. In many practical cases, our system to adapt and react to changes in noise conditions interest lies in determining the location of the dominant rapidly. This real-time data capture, combined with noise source, its level, and its contribution to the total efficient data management practices, enables us to noise level. Due to the random statistical properties of assemble a comprehensive noise source map. These maps environmental noise sources, long time intervals are are capable of not only identifying and classifying different required for such an assessment. After the equivalent noise sources but also tracking their temporal variations. level stabilizes within this time interval of duration τ 0, we Hence, our methodology offers an effective means to can classify the entire acoustic system as stable within the identify, localize, and classify noise sources in real-time, length τ 0 of the measured time interval. Arguably, providing a more accurate and comprehensive view of the integrated samples describe the population of noise noise environment in any given area. These noise source events and hence, of noise sources. maps, updated in real time, are powerful tools that can contribute to more efficient control of noise pollution. Let's observe waveforms during an arbitrary time window of duration τ 0. The contribution of all noise sources can be easily summed up at the arbitrary immission point IP j. The 2.1 Immission directivity contribution of each source W i at the selected point is p RMS,j,i and the effective overall sound pressure at IP j is Let us assume an emission plane with a finite number of p RMS,j can be calculated by a simple equation, Eq.1. noise sources 𝑊𝑖. These noise sources can be simplified and represented as monopole sources, each emitting an arbitrary wave function. The radiated noise from these Prezelj et al.: Immission Directivity as a tool for generation of Noise Maps 133 AAAA – 2023 – IZOLA - Conference Proceedings 𝑀 immission directivity measured on two locations, 𝑝2 2 assuming that noise sources are monopole sources. The 𝑅𝑀𝑆,𝑗 = ∑ 𝑝𝑅𝑀𝑆,𝑗,𝑖 (1) challenge arises as to whether it is possible to use the i=1 immission directivity in more complex sound fields with Here, the total effective sound pressure 𝑝2 more noise sources than there are available immission 𝑅𝑀𝑆,𝑗 at the selected immission Point (IP) is the sum of all M sound points, as shown in Figure 4. waves propagating from all surrounding directions (noise sources on the parallel emission plane) towards the IP (elements on the immission plan). The discretization from the random arrangement of M sources around the immission point in the Cartesian plane (Fig.2, left) can be reconfigured into the radial coordinate system (Fig.2, right). This reconfiguration allows us to define the Fig. 3: Simulated Sound pressure on the immission plane immission directivity. and immission directivity at two selected immission points. Two sound sources are located on the emission plane. Fig. 2: Superimposed immission plane over the emission Fig. 4: Simulated sound pressure on the immission plane planes at height h in cartesian and in radial coordinates. and immission directivity at two selected immission Wi represents sources with sound power W>0 and IP points. 32 sound sources with random power are represents selected Immission point. randomly distributed on the emission plane. The sound pressure level at the selected IP j is a sum of all M sound waves arriving at the IP, as shown in Fig.2 left, and described in Eq.1. Therefore, we can rewrite the 2.2 Differential microphone array equation Eq. 1 for effective sound pressure at the IP for The process of measuring immission directivity the radial coordinate system and rewrite it in a more necessitates the use of a microphone array and a general form as the integral of contributions of sound beamforming algorithm, tools that aid in localizing sound pressure from all directions in the plane. This form is used sources or determining the Direction Of Arrival (DOA) of for the definition of sound immission directivity p RMS(). sound. Differential microphone arrays (DMAs), a distinct subset in sound localization, have emerged as the most 2𝜋 2𝜋 appropriate beamforming method for a variety of 𝑝2 2 2 2 𝑅𝑀𝑆 = ∑ 𝑝𝑅𝑀𝑆,φ ⇒ 𝑝𝑅𝑀𝑆 = ∫ 𝑝𝑅𝑀𝑆(𝜑) 𝑑𝜑 (2) applications that require speech recognition over the past 𝜑=0 0 decade. These applications include hands-free systems, In Figure 3, in the middle, the sound pressure immission mobile phones, and hearing aids. Circular DMAs have plane is shown, parallel to the sound emission plane and undergone extensive examination for speech and audio uniformly distanced h above it. Two points i.e., the two applications, primarily due to their adaptability in control, sound sources can be clearly identified by bright orange the capacity to establish frequency-invariant directivity colour for higher values of sound pressure, along with how patterns, and superior directivity factor [35]. Their the sound pressure decays with distancing from the inherent benefits include significant beam alignment, source, where the sound level drops by 6 dB per doubling frequency independence, and a compact geometrical of the distance. The figure 3 also indicates the two configuration [36]. Several models have been constructed immission points (white dots in a circle) and the immission where all the first-order DMA computations are directivity p conducted in the time domain. This approach offers a RMS() in these two points, Figure 3 left and right. The result clearly illustrates the ease of determining crucial advantage - the computations exhibit minimal the location and sound power of two sources from latency. This feature is particularly vital and advantageous Prezelj et al.: Immission Directivity as a tool for generation of Noise Maps 134 AAAA – 2023 – IZOLA - Conference Proceedings in real-time applications [37-39]. The principle of Using a differential array with four microphones and a amalgamating microphone signals from a circular array diameter/side length of 𝐷𝑎𝑟𝑟𝑎𝑦 = 0.04 m, it is possible to [43] sparked the development of our unique algorithm shorten the length of the direction detection sub-window [40-42]. This algorithm, designed to compute the DOA, to 2 ∙ 𝐷𝑎𝑟𝑟𝑎𝑦⁄𝑐 ≈ 0.232 ms. Shortening the subwindow utilizes a simplified method based on timed delay to 0.232 ms provides 44 signal values for calculating DOA cascading pairs (as depicted in Figure 4 and given by Eq. of the dominant sound source. If the signals from two (3)) to form a DMA. The differential beamforming microphones placed 40 mm apart are recorded at a algorithm based on sub-windowing (SubW-DBA), is sampling rate of 192 kHz, then 22 samples are needed to developed for this application and is described in our arrange the signals for the maximum delay for each previous work [44]. SubW-DBA has an advantage in that it direction, and thus the time resolution of detecting the is not limited to the low-frequency range, at least dominant direction is 0.232 ms, [44]. theoretically. In practice, the low-frequency limit is defined by the phase matching of the microphone pairs. The high-frequency limit depends on the nature of the sound. If the sound is random, there is no frequency limit. If the sound is harmonic, the boundary is at the distance between the microphone pairs, like the usage of sound intensity probes. The algorithm defined in Eq. (3) was used to compute the instantaneous DOA to experimentally verify the hypothesis that the 2D immission directivity pattern can be obtained by associating the instantaneous total sound level (SPL) with the instantaneous dominant direction (DOA) during sub-windowing. The objective is to compile a collection of immission vectors from which an immission directivity pattern can be derived through consistent Fig.5: Differential microphone array of the first order, integration/averaging of the immission vectors. used during the experiment. Two pairs were used for the calculation of DOA on the immission plane, [44]. 𝐷 𝑁 𝐷 2 𝑎𝑟𝑔𝑚𝑖𝑛 𝑗 = ‖∑ [𝑥mic1(𝑛) − 𝑥mic2 (𝑛 − + 𝑖)] ‖ (3) 2 𝑖=0 𝑗=1 2.3 Noise source mapping x Noise source mapping or sound source localization, a key mic,1(n) and x mic,2(n) represent in Eq.3 the signals from microphone 1 and microphone 2 respectively. N denotes procedure in fields such as environmental sciences, the number of directions. D telecommunications, and noise control, is defined by its min is the number of samples in the observed window. The minimum value of D ability to identify and quantify noise sources within a min is defined by the speed of sound c, the sampling frequency specific domain or structure. The propagation of sound f from a noise source to a receiving plane, as well as the s and the distance between two microphones X mic, which determine the time resolution of direction detection. relationships between sound power, sound pressure, and Index i represents the steering value, i.e., Δn, which is distance from the sound source are fundamental to this proportional to the time delay Δτ. process. 𝑋 𝑝2𝑅𝑀𝑆 𝐷 𝑚𝑖𝑐 𝑓𝑠 (4) 𝑊 = ∮ 𝑝𝑣𝑑𝑆 = ∮ 𝑑𝑆 (6) 𝑚𝑖𝑛 = 𝑐 𝜌𝑐 For the fastest possible time response of detecting the Eq.(6) describes the relationship between sound power direction we can set 𝑀 = 𝑁. W, sound pressure p RMS, and acoustic volume flow through the surface S in a free acoustic field and in the far 𝐷 field of a noise source. By integrating across the surface of ∆𝑡 𝑚𝑖𝑛 𝑚𝑖𝑛 = (5) 𝑓𝑠 a sphere S, we derive Eq.(7) for sound power, which highlights the link between sound power W and sound pressure p RMS at a distance r. Prezelj et al.: Immission Directivity as a tool for generation of Noise Maps 135 AAAA – 2023 – IZOLA - Conference Proceedings where indices m and n run along the source plane and 2𝜋𝑟2 𝑊 = 𝑝2 indices i,j along the receiving plane. 𝑅𝑀𝑆 (7) 𝜌 𝑐0 𝜌 𝑐0 In the scenario where a monopole source is positioned on 𝑨𝑻𝑴 = 2 (12) 2(1 + 𝛼)𝜋 |𝑅 a plane with an absorption coefficient α, only a portion of 𝑚𝑛 − 𝑟𝑖𝑗 | the sound source energy radiates into a hemisphere. If the sound pressure 𝒑 ⃑⃑ on the receiving plane and the Eq.(8) describes how to calculate the amplitude of sound geometry of the source plane is well defined, then a vector pressure p RMS at a distance r from a sound source with power W on the emission plane with absorption 𝑾 ⃑⃑⃑ and hence the noise source maps can be calculated coefficient . This equation assumes that each element on backward from Eq. (13) using the inverse of the Acoustic the sound source plane is a monopole source on a partially Transfer Matrix (ATM). absorbing surface. 𝑾 ⃑⃑⃑ = 𝑨𝑻𝑴−𝟏 ∙ 𝒑 ⃑⃑ (13) 𝜌 𝑐 𝑝2 0 𝑅𝑀𝑆 = ( ) 𝑊 (8) 2(1 + 𝛼)𝜋𝑟2 However, the practicality of this approach is limited as it demands a large number of measurements across the The receiving plane is of the same dimension as the entire receiving plane. To circumvent this limitation, transmitting plane, denoted as M x N. Every element on immission directivity measurements can be employed. the immission plane is hit by a sound wave emitted from The construction of a comprehensive sound source map all the constituent elements on the emitting plane, each necessitates the acquisition of an ample amount of data. of which represent the noise source as monopoles. In this Under ideal conditions, the measurement of sound scenario, those elements on the sound emitting plane pressure at a select few immission points can provide closer to an element on the receiving plane contribute sufficient data for theoretical models, as depicted in Fig. more significantly than the distant elements. This 3. However, in real-world scenarios, as illustrated in Fig. 4, relationship can be represented by Eq.9 in which 𝑟⃑ the construction of an accurate noise map requires the epresents the vector the receiving element and 𝑅⃑ signifies extraction of substantially more information from the the vector to the sound source. The structure of this measurements. equation bears similarity to the Rayleigh integral used in Immission directivity measurements provide sufficient vibroacoustics. Furthermore, the Eq.9 can be information about the sound field in a specific location, reformulated in discrete domain given by Eq.10, in which thereby facilitating the creation of a sound source map. r ij represents the radius to the point on the receiving plane The formulation of immission directivity, as defined in Eq. and R mn signifies the radius to the point on the 2, can be extended further. The contributions of all sound transmitting plane, as depicted in Fig. 2. sources at various distances and at angle  need to be integrated as per Eq. 14. Here, W(, r) represents the 𝜌 𝑐 𝑊 spatial distribution of sound power along the radius r from 𝑝2 0 𝑅𝑀𝑆 (𝑟) = ∬ 𝑑𝑥𝑑𝑦 (9) 2(1 + 𝛼)𝜋 2 the immission point at angle , demonstrated on the right |𝑅⃑ − 𝑟⃑| 𝑥,𝑦 of Fig. 6. 𝑀 𝑁 𝜌 𝑐 𝑊 ∞ 𝑝2 0 𝑚,𝑛 𝜌 𝑐0 𝑊(𝜑, 𝑟) 𝑅𝑀𝑆,𝑖𝑗 = ∑ ∑ (10) 2 2(1 + 𝛼)𝜋 𝑅2 2 𝑝 (𝜑) = ∫ 𝑑𝑟 (14) 𝑚,𝑛 − 𝑟 𝑅𝑀𝑆 𝑖,𝑗 𝑚=1 𝑛=1 2(1 + 𝛼)𝜋 𝑟2 𝑟=𝑟0 Equation can be rewritten in the matrix form: 𝑅𝑚𝑎𝑥 𝜌 𝑐 𝑊(𝜑, 𝑟) (15) 𝑝2 0 𝑅𝑀𝑆(𝜑) = ∑ 𝒑 ⃑⃑ = 𝑨𝑻𝑴 ∙ 𝑾 ⃑⃑⃑ (11) 2(1 + 𝛼)𝜋 𝑟2 𝑟=𝑟0 Where components of the vector 𝒑 ⃑⃑ represent all elements If the function W(, r) was known, we could directly 𝑝2𝑅𝑀𝑆,𝑗 on the receiving plane and 𝑾 ⃑⃑⃑ all sound sources perform noise source mapping from it. However, in most elements W i on the sound emitting plane. Both vectors instances, this function remains elusive. Nevertheless, we have dimension M∙N. ATM is an Acoustic Transfer Matrix possess knowledge of its integral value, defined within Eq. with dimensions M∙N x M∙N and is defined in Eq. (12) by 14 i.e., Eq.15 in terms of the contribution of sound pressure from the direction . Leveraging this Prezelj et al.: Immission Directivity as a tool for generation of Noise Maps 136 AAAA – 2023 – IZOLA - Conference Proceedings information, we can scrutinize a typical scenario of spatial 2(1 + 𝛼)𝜋 (16) distribution of environmental noise sources (denoted by 𝑊(𝜑, 𝑟) = 𝑝2𝑅𝑀𝑆(𝜑) 𝑟2 red squares) and emission point, as shown on Fig. 6 above. 𝜌 𝑐0 It follows that we can write a vector 𝑾 ⃑⃑⃑ with the dimension equal to the number of immission points with components 𝑾 ⃑⃑ = [W (1, r), W (2, r), …. W (N, r)]. A simple statistical analysis can be performed on the components of the vector and the true sound power value for the observed sound source W can be extracted from it. During the measurement phase, we can also establish the lower limit of integration r 0 by estimating the distance to the nearest real noise source, allowing for further improvement of noise source mapping. 3. EXPERIMENTAL RESULTS To conduct immission directivity measurements, we employed a differential microphone antenna (DMA) equipped with four measurement microphones. The digitization of each signal was performed at a sampling rate of 192 kHz with a 24-bit resolution. Prior to experimentation, the microphones underwent calibration and phase compensation. Signal acquisition was facilitated by LabView software, which enables the real- time determination of the Direction of Arrival (DOA) at 50- millisecond intervals. The immission directivity was derived through the process of energy averaging over a Fig.6: Typical Environmental Noise Source Configuration ten-minute duration at each immission point. and Measurement Locations of Immission Directivity Depicted are the noise sources (represented by squares), along with the points IP of immission directivity measurements (above), and a plot of sound power along the radius at selected angles, extending from the ( IP j) (below). Observing a single source W (represented by the black square), we perceive other sound sources (red squares) as parasitic. On the right of Fig. 6, the functions W(, r) for Fig. 7: A) Prototype system using DMA with square four distinct immission locations are illustrated, through geometry and B) System for microphone calibration and which we ascertain the sound power of the source W. phase matching. Upon analysing these functions, it is evident that there is always a probability of encountering multiple parasitic The experiments were carried out under real-world sound sources for any chosen direction of immission, as conditions at four distinct locations: • illustrated in the function W ( Industrial premises containing a carpentry. 1, r) for IP1, denoted with • black curve. However, the likelihood of encountering Surrounding of a busy traffic junction • several parasitic sound sources across all functions; W ( Industrial premises of oxygen plant in Kranj 1, • r), W ( Industrial premises of Sandmilll Kresnice 2, r), W (3, r), and W (4, r), aimed at determining the sound source W is exceedingly small. Because function These varied locations were selected to provide a diverse is discrete, with only one peak, its integral value can be range of environmental conditions and sound sources. simply attributed to the elements on the emitting plane with the distance r, as given by Eq.16. Prezelj et al.: Immission Directivity as a tool for generation of Noise Maps 137 AAAA – 2023 – IZOLA - Conference Proceedings 3.1 Noise mapping of industrial premises - Carpentry In summary, The ventilation system always emerged as the dominant noise source, with other contributors being Our inaugural measurement set was conducted in the a neighbouring workshop, a stationary truck, and vicinity of Ovsenik Carpentry in Britof, henceforth referred agricultural activity. Although several measurements did to as 'the carpentry.' The primary noise contributors in not align with our initial expectations due to unforeseen this locale are the woodworking machinery in the yard and noise sources, the collected data were adequate to the workshop's ventilation system. We conducted identify and spatially determine the dominant noise immission directivity measurements at nine points (IP1- sources at the carpentry. IP9) encircling the entire complex. Several factors affected the clarity of the results at some locations, notably IP4 and IP5. The carpentry's surroundings comprise fields and forests, contributing ambient noise from tree felling and agricultural machinery operation. Furthermore, a nearby highway intermittently added to the ambient noise during relatively quiet periods. In the measurements at IP1, the primary noise source at 225° and 81 dB was evidently the ventilation system. Immission directionality measured at IP2 Fig. 8: Measurement results of immission directivity at continued to show the ventilation system as the dominant locations IP1, IP3, IP4, IP 5 and IP9, around the carpentry noise source, although the peak was less distinct. Noise with coordinates on the map and synthesized individual contributions from the neighbouring farm and the sound power maps predicted for each measurement carpentry's machinery were also included. For IP3, the location. ventilation system noise again dominated at 162° and 77 dB. However, the workshop's yard machinery noise was not detected due to obstruction by a building. The measurement at IP4 differed significantly from previous readings. Here, the dominant noise source was not as noticeable, with the largest contribution coming from the carpentry yard at 108° and 75 dB. However, noise from forest sawing and nearby tractors was also perceptible. At IP5, multiple noise sources ranging from 45° to 200° were observed, originating from the carpentry yard's machinery. However, the dominant noise source was a stationary truck at 300° and 76 dB. Although we initially anticipated the machinery to be noisier than the truck, a Fig.9: Combined Sound power map of the carpentry, yard fence interfered with the readings. The IP6 synthesized from 5 sound power maps, for locations IP1, measurement introduced a new ventilation system, the IP3, IP4, IP 5 and IP9 dominant noise source at 108° and 81 dB. Other significant noise contributors were work in a nearby field (354°, 66 dB), a stationary truck (24°), and forest sawing 3.2 Noise mapping of traffic junction (60°). At IP7, despite our initial expectation of prevailing noise from tractor work in nearby fields due to our Our objective was to assess noise pollution originating increased distance from the carpentry, the yard work from traffic, one of the key noise producers in an urban noise still dominated. The leading source was recorded at environment. Unlike our previous location, where 84° with significant amplitude. For IP8, as expected, the stationary sources like a ventilation system workshop's ventilation system was the dominant noise predominated, we anticipated fewer distinct peaks here source, discernible at 114° with an amplitude of 74 dB. due to the transient nature of vehicles. The junction, The IP9 measurements were unclear, although the situated near the Brnik airport, facilitated a heavy flow of conditions were like the first measurement. Despite our both cars and trucks, given the presence of several expectations, the maximum amplitude measured was at warehouses in the vicinity. We conducted nine 60° and 66 dB, which was less evident from the graph. measurements in parallel with the two main road encompassing the junction. Prezelj et al.: Immission Directivity as a tool for generation of Noise Maps 138 AAAA – 2023 – IZOLA - Conference Proceedings The results of the immission directivity measurement at IP 3.3 Noise mapping of industrial premises - oxygen plant 1, with an average amplitude of 71 dB, were directed Our third test location featured two constant noise towards the road, as predicted. There were no extraneous sources - the Škofja Loka oxygen plant and the adjacent noise sources apart from vehicular traffic, providing an Urbanscape factory. While our focus was primarily on the accurate representation of the noise immission. Similarly, oxygen plant, measurements 1, 2, and 8 incidentally IP2 measurements also pointed towards the road with a included noise from the nearby factory. It's worth noting maximum amplitude of 81 dB. The primary noise source that measurements 1, 2, and 3 recorded the sound of a at IP 5 was the intersection, aligning with our expectations passing train, though these results could be disregarded due to our proximity to the junction. There was some as irrelevant to our study. The auditory influence of the additional noise from the road diverging from the train was evident up to the 5th measurement. intersection, but the average noise level exceeded 70 dB, as expected. However, the measurements at IP 7 and IP 4 were affected by background noise from agricultural work, but if we disregard this, the maximum amplitudes facing the road were 75 dB and 74 dB respectively. Comparing the maxima of the individual measurements illustrated that the noise originated predominantly from the traffic or the intersection. The loudest noises were detected before the intersection or road junctions, likely resulting from vehicular braking. Given the heavy traffic, Fig. 12: Measured of immission directivity at IP 1, IP2, vehicles were frequently stationary near our IP4, IP5, and IP7, with coresponding sound power maps measurement sites. These stationary, constant noise predicted for each measurement location. sources often dominated our ambient noise measurements. Fig. 10: Measurement results of immission directivity with coordinates on the map and synthesized individual Fig.13: Combined Sound power map synthesized from sound power maps predicted for each measurement 5 sound power maps IP1, IP2, IP4, IP5, and IP7 location. 3.3 Noise mapping of industrial premises - Sand mill Kresnice Our fourth site, situated at Kresnice, was characterized by the noise generated from a sand mill, with the objective of observing its noise propagation at various points. This location presented additional sources of noise which included a railway station with frequently passing trains, a bustling road across the river, and heavy machinery operating adjacent to the mill in the sand pit. As a result, apart from the main source (the sand mill), we also Fig.11: Combined Sound power map synthesized from 5 measured three ancillary noise sources. However, it sound power maps, generated from measurements of should be noted that a calibration issue with the immission directivity at IP2, IP4, IP5, IP7 and IP8 microphones affected the measurement of accurate Prezelj et al.: Immission Directivity as a tool for generation of Noise Maps 139 AAAA – 2023 – IZOLA - Conference Proceedings pressure level values. This calibration discrepancy did not This achievement has significant implications for the impact the directivity measurements. field of acoustical engineering and environmental noise management. It paves the way for more efficient noise monitoring and control, aiding urban planning, industrial site management, and other areas where noise impact is a critical factor. Further, it validates the effectiveness of utilizing immission directivity measurements as a method for noise source identification and mapping. Presented proof of concept presents the next step in the modelling of noise maps, where it is no longer just an Fig. 14: Measurement results of immission directivity extrapolation of measurements with the help of models, around sand mill Kresnice, at IP2, IP4, IP5, IP6 and IP 7, but purely based on the measurement methods itself with coordinates on the map and synthesized individual (DOA, immission directivity, etc.) sound power maps predicted for each IP location. Going forward, the demonstrated concept could be further refined, and its accuracy improved with enhanced calibration of measuring equipment. Additional studies could explore different types of noise sources, expanding the breadth of applicability of this methodology. Ultimately, the success of this proof of concept lays a strong foundation for further research and application in the realm of noise mapping based on immission directivity measurements. 4. REFERENCES [1] Environmental noise guidelines for the European region: WHO, Regional office Europe, 2018 Fig.15: Combined Sound power map synthesized from [2] Assessment of potential health benefits of noise 5 sound power maps, generated from immission abatement measures in the EU, The Phenomena directivities at IP2, IP4, IP5, IP6 and IP 7. project, Contract number 07.0203/2019/ETU/815591/ENV.A.3, March 2021. [3] Environmental noise in Europe — 2020, EEA Report 3. CONCLUSION No 22/2019 [4] Towards a Zero Pollution for Air, Water and Soil, the Through a series of systematic and varied European Commission measurements at four distinct locations, this study has [5] Luca Fredianelli, et.al, Source characterization validated the feasibility of calculating noise source maps guidelines for noise mapping of port areas, Heliyon, from immission directivity measurements. By Vol. 8, (2022), pp. e09021 characterizing different noise sources, such as industrial [6] Luca Fredianelli, et al., Classification of Noise Sources machinery, traffic at busy junctions, and stationary for Port Area Noise Mapping, Environments, Vol. equipment, we have been able to gather comprehensive 8(12), (2021), data that allow us to understand noise propagation and https://doi.org/10.3390/environments8020012 develop a visual representation of these noise sources. [7] Gwenaël Guillaumea, Pierre Aumond, Pierre Each location presented unique noise sources, and Chobeau, Arnaud Can, Statistical study of the despite certain challenges such as unexpected sources relationships between mobile and fixed stations and calibration issues, we were able to effectively capture measurements in urban environment, Building and the necessary data and achieve our objectives. Through Environment, Vol. 149, (2019), pp.404–414 meticulous analysis of this data, we were able to interpret [8] Judicaël Picaut, Arnaud Can, Nicolas Fortin, Jeremy the immission directivity measurements and thereby Ardouin, Mathieu Lagrange. Low-Cost Sensors for generate accurate noise source maps. Urban Noise Monitoring Networks-A Literature Review. Sensors, Vol.20 (8), (2020), pp. 10.3390 Prezelj et al.: Immission Directivity as a tool for generation of Noise Maps 140 AAAA – 2023 – IZOLA - Conference Proceedings [9] Catherine Lavandier, et al., Influence of road traffic [22] Maijala, P.; Zhao, S.; Heittola, T.; Virtanen, T. noise peaks on reading task performance and Environmental noise monitoring using source disturbance in a laboratory context, Acta Acustica classification in sensors, Applied Acoustics, Vol.129, (2022), Vol.6(3) Available online at: (2018), pp.129, 258–267 https://doi.org/10.1051/aacus/2021057 [23] Sevillano, X, et al. DYNAMAP—Development of low [10] Pierre Aumond, Léo Jacquesson, Arnaud Can, cost sensors networks for real time noise mapping. Probabilistic modeling framework for multisource Noise Mapping, Vol.3, (2016), pp. 172–189 sound mapping, Applied Acoustics, Vol. 139, (2018), [24] Peckens, C.; Porter, C.; Rink, T. Wireless sensor pp.34–43 networks for Log-term Monitoring of Urban Noise., [11] Weigang Wei, Timothy Van Renterghem, Bert De Sensors (2018), Vol.18, pp.3161 Coensel, Dick Botteldooren, Dynamic noise mapping: [25] Tatiana Alvares-Sanches, Patrick E. Osborne, Paul R. A map-based interpolation between noise White, Mobile surveys and machine learning can measurements with high temporal resolution, improve urban noise mapping: Beyond A-weighted Applied Acoustics, Vol. 101, (2016), pp. 127–140 measurements of exposure, Science of the Total [12] Ziqin Lan, Canming He, Ming Cai, Urban Road traffic Environment, Vol. 775, (2021), pp.145600 noise spatiotemporal distribution mapping using [26] Roberto Benocci, et al., Accuracy of the Dynamic multisource data, Transportation Research Part D, Acoustic Map in a Large City Generated by Fixed Vol. 82, (2020), pp.102323 Monitoring Units, Sensors, Vol.20, (2020), pp. 412 [13] Weijun Yang, Jinying He, Canming He, Ming Cai, [27] Roberto Benocci, et al., Optimized Sensors Network Evaluation of urban traffic noise pollution based on and Dynamical Maps for Monitoring Traffic Noise in noise maps, Transportation Research Part D, Vol. 87, a Large Urban Zone, Appl. Sci., Vol. 11, (2021), pp. (2020), pp.102516 8363 [14] B. Doohan, et.al, The sound of management: [28] G. Quintero, Balastegui, J. Romeu, Traffic noise Acoustic monitoring for agricultural industries, assessment based on mobile measurements, Ecological Indicators, Vol. 96, (2019), pp.739-746 Environmental Impact Assessment Review, Vol. 86, [15] Sacha Baclet, et al., From strategic noise maps to (2021), pp.106488 receiver-centric noise exposure sensitivity mapping, [29] G. Quintero, Balastegui, J. Romeu, A low-cost noise Transportation Research Part D, Vol. 102, (2022), measurement device for noise mapping based on pp.103114 mobile sampling, Measurement, Vol. 148, (2019), [16] Xueyi Xu, et al., Application of land use regression to pp.106894 map environmental noise in Shanghai, China, [30] Carlos Prieto Gajardo, Juan Miguel Barrigón Morillas, Environment International, Vol. 161, (2022), Guillermo Rey Gozalo, et al., Can weekly noise levels pp.107111 of urban road traffic, as predominant noise source, [17] Francesco Asdrubali, Francesco D’Alessandro, estimate annual ones?, The Journal of the Acoustical Innovative Approaches for Noise Management in Society of America 140, 3702 (2016); doi: Smart Cities: a Review, Current Pollution Reports, 10.1121/1.4966678 Vol. 4, (2018), pp.143–153 [31] Chinmayi Mahapatra, A.R. Mohanty, Optimization of [18] Alías, F.; Alsina-Pagès, R.M. Review of Wireless number of microphones and microphone spacing Acoustic Sensor Networks for Environmental Noise using time delay based multilateration approach for Monitoring in Smart Cities, Journal of Sensors, Vol. explosive sound source localization, Applied 2019,(2019), pp.7634860 Acoustics, Vol. 198, (2022), pp.108998 [19] Roberto Benocci, et al., Mapping of the Acoustic [32] Yong Chen, Ming Wua, Ling Lu, Xing Fang, Jun Yang, Environment at an Urban Park in the City Area of Sound pressure level real-time monitoring of Milan, Italy, Using Very Low-Cost Sensors, Sensors, outdoor directional noise, Applied Acoustics, Vol. Vol. 22, (2022), pp. 3528 202, (2023), pp. 109141 [20] Juan Manuel López, et al., A Digital Signal Processor [33] John P. Djungha Okitadiowo, A. Lay-Ekuakille, T. Based Acoustic Sensor for Outdoor Noise Monitoring Isernia, A. Massaro, Design of a beamforming in Smart Cities, Sensors, Vol. 20, (2020), pp.605 antenna sensor for environmental noise detection to [21] Mydlarz, C.; Salamon, J.; Bello, J.P. The discriminate vehicle emission according to road implementation of low-cost urban acoustic conditions, Measurement: Sensors, Vol. 23, (2022), monitoring devices, Applied Acoustics, Vol. 117, pp.100389 (2017), pp. 207–218. [CrossRef] Prezelj et al.: Immission Directivity as a tool for generation of Noise Maps 141 AAAA – 2023 – IZOLA - Conference Proceedings [34] Raphaël Leiba, et al., Acoustical Classification of the Array, 138th AES Convention, 2015, Warsaw, Poland, Urban Road Traffic with Large Arrays of Paper 9335. Microphones, Acta Acustica united with Acustica, [42] Todorović Dejan, Salom Iva, Čelebić Vladimir, and Vol. 105. (2019), pp.1067 – 1077 Prezelj Jurij, Implementation and Application of [35] Gongping Huang, Jacob Benesty and Jingdong Chen, FPGA Platform with Digital MEMS Microphone Array, Design of robust concentric circular differential Proceedings of 4th International Conference on microphone arrays, The Journal of the Acoustical Electrical, Electronics and Computing Engineering, Society of America 141, 3236 (2017); IcETRAN 2017, Kladovo, Serbia, June 05-08, ISBN https://doi.org/10.1121/1.4983122 978-86-7466-692-0. [36] Buchris, Y., I. Cohen, and J. Benesty. Analysis and [40] Murovec, J., Curovic, L., Novakovic, T., & Prezelj, J., design of time-domain first-order circular differential Environmental noise event classification based on microphone arrays. The 22nd International Congress self-organizing map using psychoacoustic features on Acoustics (ICA). 2016. and spatial filtering. Universitätsbibliothek der [37] Benesty, Jacob, and Chen Jingdong. Study and design RWTH Aachen, 2019. of differential microphone arrays. Vol. 6. Springer [43] Huang, Gongping, Jacob Benesty, and Jingdong Science & Business Media, 2012. Chen. Design of robust concentric circular [38] Buchris, Yaakov, Israel Cohen, and Jacob Benesty. differential microphone arrays. The Journal of the First-order differential microphone arrays from a Acoustical Society of America 141.5 (2017): 3236– time-domain broadband perspective. 2016 IEEE 3249. international workshop on acoustic signal [44] J Prezelj, L Čurovič, T Novakovič, J Murovec, A novel enhancement (IWAENC). IEEE, 2016. approach to localization of environmental noise [39] Buchris, Yaakov, Israel Cohen, and Jacob Benesty. On sources: Sub-windowing for time domain the design of time-domain differential microphone beamforming, Applied Acoustics, Vol.195, (2022), arrays. Applied Acoustics 148 (2019): 212–222. pp.108836 [41] Salom Iva, Čelebič Vladimir, Todorović Dejan and Prezelj Jurij, An Implementation of Beamforming Algorithm on FPGA Platform with Digital Microphone Prezelj et al.: Immission Directivity as a tool for generation of Noise Maps 142 DATA SELECTION FOR REDUCED TRAINING EFFORT IN VANDALISM SOUND EVENT DETECTION Stefan Grebien1, Florian Krebs1, Ferdinand Fuhrmann1, Michael Hubner2, Stephan Veigl2, Franz Graf1 1 Joanneum Research, Intelligent Acoustic Solutions, Austria 2 AIT Austrian Institute of Technology GmbH, Center for Digital Safety & Security, Austria Abstract: ing recordings of both target and background events. These Typical sound event detection (SED) applications, employed in datasets should, in the ideal case, contain a representative col- real environments, generate huge amounts of unlabeled data lection of acoustic events, that will occur within the environ- each day. These data can potentially be used to re-train the ment of the deployed system. Nevertheless, in practical terms, underlying machine learning models. However, as the labeling the acoustic environment of the deployed system is often un- budget is usually restricted, active learning plays a vital role in known during development and undergoes changes over time re-training. Especially for applications with sparse event occur-due to external factors like weather, seasons, temporal construc- rence, a data selection process is paramount. In this paper we (i) tion site noise, or other influences. Therefore, it is important introduce a novel application for vandalism SED, and (ii) analyze to update datasets and classifiers regularly in order to react to an active learning scheme for reduced training and annotation these changes and adapt the system to local conditions. This can effort. be implemented by periodically storing sound segments, select- In the presented system, the employed machine learning classi- ing the most relevant ones, and annotating them in order to re- fier shall recognize various acts of vandalism, i.e., glass breakage train the model. However, the selection and annotation process and graffiti spraying. To this end, we utilize embeddings gen- is tedious if performed manually. To this end, active learning erated with a pre-trained network and train a recurrent neural offers ways to decrease the amount of human labelling effort. network for event detection. The applied data selection strategy While there exist several active learning approaches for im-is based on a mismatch-first, farthest-traversal approach and is age processing tasks [4–6], the literature is rather sparse for au-compared to an upper bound by using all available data. Fur- dio event detection. One of the few examples is an uncertainty- thermore, results for the active learning scheme are evaluated based selection strategy for a noise detection problem [7]. An-with respect to different labeling budgets and compared to an other one is mismatch-first-farthest-traversal (MFFT) [8], a data active learning scheme with a random sampling scheme. selection method which iteratively samples new data from mis- Keywords: sound event detection, active learning, dataset se-classified segments. At the same time, diversity between data lection, vandalism items is enforced by maximizing the distance between currently and previously sampled data items. And most recently, Shishkin et al. [9] used pretrained embeddings and trained the last layer 1. INTRODUCTION with Monte Carlo dropout to identify samples with high un- Vandalism, the deliberate defacement or destruction of infras- certainty. Again, samples with high uncertainty were selected tructure, is a growing challenge to the public and causes eco- first. However, as each approach was validated using a differ- nomic damage due to the increased maintenance costs. There- ent dataset, it is still unclear which data selection strategy works fore, there is a growing demand for systems that are able to best for a given application. detect acts of vandalism in real time in order to facilitate im- Therefore, this paper aims to provide the following contri- mediate responses. Among these systems, acoustic monitoring butions: (i) We propose a new system for acoustic vandalism emerges as a promising avenue, offering a non-intrusive means detection which recognizes the sounds of glass-breaking, spray- to detect and characterize acts of vandalism in various settings. ing, and shaking a graffiti can. (ii) We apply and analyze MFFT Acoustic systems have therefore been proposed to detect events on both synthetic and real-world datasets in order to select the related to vandalism such as shouting [1], kicking-objects [2], most relevant datapoints for training the vandalism detector. screaming [2] and glass breaking [3]. With these contributions, this work aims to take another step At the core of these acoustic monitoring systems lie clas- towards a practical applicability of active learning methods for sifiers, typically trained on annotated datasets encompass- SED systems. Grebien et al.: Data Selection for Reduced Training Effort in Vandalism Sound Event Detection 143 2. PROPOSED SYSTEM between model-predicted and model-propagated labels, with mismatching segments coming first. The proposed system closely follows the active learning scheme presented in [8]. The input are unlabeled recordings and the 2. Farthest traversal: This criterion aims at increasing the output is a classifier for SED. At the beginning of the iterative diversity within the selected data. It selects the sample active learning scheme, a subset of the data is selected to train farthest away from the already selected samples, using an SED classifier. In the subsequent iterations, the trained SED the ordering described above. To this end, the cosine dis- classifier is utilized by the data selection process to generate the tances between the mean of embeddings of the labeled next training subset. In the following subsections we will present and unlabeled data are computed. The data selection an overview of the system and highlight the differences to [8]. scheme adds one sample at a time, to ensure that the chosen sample is farthest away from the labeled data and the already selected samples in this iteration. Note that 2.1 Data pre-processing while mismatch-first can only be applied after the first As suggested by Shuyang et al. [8], the system utilizes a pre- iteration, farthest traversal is already used to select the trained network to generate embeddings. This network follows first batch, except for the very first sample which needs the architecture in [10], the training material and validation cri-to be chosen randomly. terion generally follows [11] and the trained embedding network is available online [12]. The input to the embedding network are The newly selected samples are then presented to an annotator log-mel spectrograms with who needs to assign weak labels to the chosen segments. By 128 mel-bands, produced from raw audio recordings with a sampling frequency of requiring weak labels only, the annotation effort can be reduced 44.1 kHz, a hop- size of even more. 882 samples and a frame-size of 1764 samples. The em- bedding network produces an output of The samples selected for training are used to train to a bi- 256 embeddings at an output rate of directional recurrent neural network (RNN) consisting of three 50 Hz. In order to split the files into shorter, homogeneous seg- layers of gated-recurrent units (GRUs) with 30 hidden units each. ments, each file is fed into a change point detection module, The model outputs y for each time-step. A fully connected sig- t where change points are inferred from the peaks of a change moid layer is used to generate class probabilities p for each t function. This change function is computed for each time frame time step. The weak label output used during training is com- by the cosine distance of the mean over the puted in a similar fashion as in [8]. As we analyze binary classi- K past embeddings and the mean over the fiers only, we apply a sigmoid instead of a softmax function for K future embeddings, where K is set in such a way that generating the attention output. 2K corresponds to 1 sec. While [8] uses full recordings as input to the change point detection module, our system splits the raw audio recordings first in 10 sec long 3. EXPERIMENTS segments. 1 The change point module returns shorter segments with a minimum length of 1 sec which are used as input for the We employ the proposed active learning scheme on three binary data-selection scheme in the active learning process. classifiers. The three different classes are glassbreaking ( glassbreak), spraying ( spray), and shaking a graffiti can ( rattle). As we aim at analysing the three classifiers independently, we train 2.2 Active learning scheme three individual classifiers in this work, instead of one mulilabel The data selection process uses MFFT to choose a subset of the classifier. preprocessed segments, also called samples. MFFT utilizes two techniques for sample selection: 3.1 Datasets 1. Mismatch first: After the first iteration, i.e., as soon as a trained classifier is available, the unlabeled samples are In order to evaluate the proposed vandalism detection system, evaluated with the model to generate two datasets are used. A summary is given in Tab. 1. model-predicted labels. Furthermore, The first one, which is used for training the glassbreak classi- model-propagated labels are calcu- lated using a nearest-neighbor classifier, i.e., the label of fier, is a subset of the TUT Rare Sound Events 2017 set [13]. It was a unlabeled segment is predicted by finding the small-generated by mixing the target sound events from Freesound est cosine distance between the means of the embed- with background sounds from the TUT Acoustic Scenes 2016 dings of the labeled and unlabeled data. Then, the set of dataset [14] in various event-to-background ratios. As we are unlabeled samples is ordered according to the mismatch interested in vandalism detection, we only use the glassbreak subset in this paper. In contrast to the setup in [8], we partition 1 This is necessary, as we do not only use synthetic data, where all the dataset into non-overlapping 10-second chunks, in order to the recordings have the same length, e.g., data from the TUT Rare Sound match the format of the second dataset. This yields 1500 seg-Events 2017 [13], but also real-world data with durations of up to 20 min. ments of the devtrain split for training, and 1500 segments of Grebien et al.: Data Selection for Reduced Training Effort in Vandalism Sound Event Detection 144 the devtest for testing. Approximately one sixth of the segments average of the F1 score as well as the minimum and maximum contain a target event. of the average F1 score with respect to the training iterations. The second dataset is an inhouse-dataset which consists of Clearly, MFFT data selection ( ) adds a disproportion- three parts of approximately equal size: (i) recordings of spray ately high number of positive samples to the training data during and can rattle sounds in various acoustic environments, (ii) falsethe first few iterations of the active learning scheme. The pos- positive detections recorded at various installations of the Graf- itive samples chosen by random data selection ( ) are (as fiti BusterTM 2 , and (iii) ambience recordings from various loca- expected) directly proportional to the current percentage of the tions in the city center of Graz, Austria. The recordings are cut training dataset size. into non-overlapping segments of 10 seconds each, and were For the glassbreak dataset, it is possible to achieve a F1 score manually annotated. of about 95% of the maximum F1 score already at a training dataset size of about 7 % with MFFT data selection. When using Total Ambient Target random sampling ( ), about 20 % of the dataset is needed [minutes] [minutes] [minutes] to achieve this performance. Glassbreak train 250 0 6 For both the rattle and the spray dataset, MFFT ( ) out- Glassbreak test 250 0 3 performs random sampling ( ) for all dataset sizes except Spray train 1003 250 36 for 1 %. However, 95% of the maximum performance is only Spray test 217 0 17 achieved when using at least 20 % of the data. We conjecture that these two datasets do not contain enough redundancy for Can rattle train 498 250 17 the algorithm to choose a near optimal subset at lower percent- Can rattle test 235 0 11 ages. This could be explained by the fact, that the datasets pre- sented in Tab. 1 are already a selected subset of the sound scene Table 1. Summary of the datasets used in this paper. at a sensor, containing mainly true positive and false positive examples of a previously trained classifier. Most of the redundant data at a specific installation is therefore already filtered out be-Note that the spray and rattle datasets are already pre-forehand. selected data used for training and do not represent the real sparsity of the events. 4. CONCLUSION 3.2 Results In this paper we introduced a novel application for sound event To be able to assess the performance of the data selec- detection namely vandalism detection and analyzed an active tion scheme, we apply the whole active learning scheme learning scheme for reduced training and annotation effort. The for each classifier three times (i.e., three presented active learning scheme selects data on the basis of selection itera- tions). The analyzed sizes of the subset used for training are a mismatch-first farthest-traversal criterion and clearly outper- forms random sampling. Furthermore, we have shown that [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 12, 14, 16, 18, 20, 60, 100] %. In order to assess the influence of the (stochastic) training process on good performance can be achieved when using about 20 % of the performance, three networks are trained for each selected the datasets. training subset, and the mean of the results among these three The next steps are to employ the active learning to the com- plete datasets for rattling and spraying a graffiti can. Addition-training iterations are presented. Note that for MFFT, we always used the third (of the three trained models) to select the data ally, we want to compare the effect of employing the algorithm for the next selection iteration, using a threshold of to balanced and unbalanced multi-label datasets. 0.5. During training, 30 % of the selected training batch is randomly chosen as validation set. 5. ACKNOWLEDGEMENT In this study, the performance of the trained classifiers are The support of the Austrian Federal Ministry for Climate Ac- evaluated with a segment-based F1 score with segment-length tion, Environment, Energy, Mobility, Innovation and Technology chosen to be 1sec [15]. For each individual result, the threshold in the program leading to the best Mobility of the Future within the project MOBI- F1 score is chosen. The results for the three different event classes are pre- LIZE (38924375) is gratefully acknowledged. sented in Fig. 1. The upper row depicts the percentage of posi- tive samples within the current selection versus the dataset size, 6. REFERENCES in percent of the total available data. The lower row shows the [1] P. W. van Hengel and T. C. Andringa, “Verbal aggression de- 2 tection in complex social environments,” in IEEE Conference https://www.joanneum.at/en/digital/ productssolutions/graffiti-buster accessed on on Advanced Video and Signal Based Surveillance, pp. 15–20, 25.08.2023 IEEE, 2007. Grebien et al.: Data Selection for Reduced Training Effort in Vandalism Sound Event Detection Grebien et al.: Data Selection for Reduced Gr Training ebien et Effort al.: Dain ta Vandalism Selection fSound or R Even educed Gr tTDe r t ebienection aining et Effort al.: Dain ta Vandalism Selection fSound or R Even educed tTDe r tection aining Effort in Vandalism Sound Event Detection Grebien et al.: Data Selection for Reduced Gr Training ebien et Effort al.: Dain ta Vandalism Selection fSound or R Even educed Gr tTDe r t ebienection aining et Effort al.: Dain ta Vandalism Selection fSound or R Even educed tTDe r tection aining Effort in Vandalism Sound Event Detection 145 100 100 100 % % % in in in samples samples samples ve ve ve 10 10 10 positi positi positi percentage percentage percentage 1 1 1 1 10 100 1 10 100 1 10 100 percentage in % percentage in % percentage in % (a) glassbreak (b) rattle (c) spray 0.8 0.8 0.8 ⟩ 1 ⟩ 1 ⟩ 1 ⟨F ⟨F ⟨F 0.6 0.6 0.6 0.4 0.4 0.4 1 10 100 1 10 100 1 10 100 percentage in % percentage in % percentage in % (d) glassbreak (e) rattle (f) spray MFFT Random Fig. 1. Results of the active learning scheme for MFFT-selection ( ) and random-selection ( ): The upper row shows the percentage of positive samples selected by the data selection scheme as a function of the overall percentage of the dataset. The lower row presents the average F1 -score including minimum and maximum. The three columns correspond to the different event classes: (a), (d) are for glassbreak, (b), (e) are for rattle, (c), (f) is for spraying. [2] J. F. Kooij, M. C. Liem, J. D. Krijnders, T. C. Andringa, and D. M. Proceedings of the IEEE/CVF conference on computer vision Gavrila, “Multi-modal human aggression detection,” Com- and pattern recognition, pp. 93–102, 2019. puter Vision and Image Understanding, vol. 144, pp. 106– [6] P. Ren, Y. Xiao, X. Chang, P.-Y. Huang, Z. Li, B. B. Gupta, 120, 2016. X. Chen, and X. Wang, “A survey of deep active learning,” [3] P. Foggia, N. Petkov, A. Saggese, N. Strisciuglio, and M. Vento, ACM computing surveys (CSUR), vol. 54, no. 9, pp. 1–40, “Reliable detection of audio events in highly noisy environ- 2021. ments,” Pattern Recognition Letters, vol. 65, pp. 22–28, 2015. [7] Y. Wang, A. E. M. Mendez, M. Cartwright, and J. P. Bello, [4] K. Wang, D. Zhang, Y. Li, R. Zhang, and L. Lin, “Cost-effective “Active learning for efficient audio annotation and classifica- active learning for deep image classification,” IEEE Transac- tion with a large amount of unlabeled data,” in IEEE Interna- tions on Circuits and Systems for Video Technology, vol. 27, tional conference on acoustics, speech and signal processing no. 12, pp. 2591–2600, 2016. (ICASSP), pp. 880–884, IEEE, 2019. [5] D. Yoo and I. S. Kweon, “Learning loss for active learning,” in [8] Z. Shuyang, T. Heittola, and T. Virtanen, “Active learning for sound event detection,” IEEE/ACM Transactions on Audio, Grebien et al.: Data Selection for Reduced Training Effort in Vandalism Sound Event Detection Grebien et al.: Data Selection for Reduced Training Effort in Vandalism Sound Event Detection 146 Speech, and Language Processing, vol. 28, pp. 2895–2905, [12] Z. Shuyang, “Audio embedding segmentation.” 2020. https://github.com/zhao-shuyang/audio_ [9] S. Shishkin, D. Hollosi, S. Doclo, and S. Goetze, “Active learn-embedding_segmentation, 2020. ing for sound event classification using monte-carlo dropout [13] A. Mesaros, T. Heittola, A. Diment, B. Elizalde, A. Shah, E. Vin-and pann embeddings,” in Workshop on Detection and Clas- cent, B. Raj, and T. Virtanen, “DCASE 2017 challenge setup: sification of Acoustic Scenes and Events (DCASE), 2021. Tasks, datasets and baseline system,” in Proceedings of the [10] Y. Xu, Q. Kong, W. Wang, and M. D. Plumbley, “Large-scale Detection and Classification of Acoustic Scenes and Events weakly supervised audio classification using gated convolu- Workshop (DCASE), 2017. tional neural network,” in 2018 IEEE International Confer- [14] A. Mesaros, T. Heittola, and T. Virtanen, “Tut database for ence on Acoustics, Speech and Signal Processing (ICASSP), acoustic scene classification and sound event detection,” in pp. 121–125, 2018. European Signal Processing Conference (EUSIPCO), pp. 1128– [11] A. Kumar, M. Khadkevich, and C. Fügen, “Knowledge transfer 1132, IEEE, 2016. from weakly labeled audio using convolutional neural net- [15] A. Mesaros, T. Heittola, and T. Virtanen, “Metrics for poly- work for sound events and scenes,” in 2018 IEEE Interna- phonic sound event detection,” Applied Sciences, vol. 6, tional Conference on Acoustics, Speech and Signal Process- no. 6, 2016. ing (ICASSP), pp. 326–330, 2018. Grebien et al.: Data Selection for Reduced Training Effort in Vandalism Sound Event Detection 147 EXPERIMENTAL SOUND FIELD CHARACTERIZATION WITH AUTOMATED HIGH- RESOLUTION IMPULSE RESPONSE MEASUREMENTS Rok Prislan1, Alexis Dufour2 1 InnoRenew CoE, SI-6310, Izola 2 CESI: École d’ing énieurs, Nanterre, France Abstract: standardized test facilities in acoustics whose requirements are The technical development of the equipment used for acous- under revision [3] - partly because of the issues related to insuf-tic measurements has allowed us to automate measurements ficient diffuseness. and to use a large number of microphones simultaneously. This To further investigate the diffuse sound field, new experi- opens up the possibility of scanning the sound field, i.e., captur-mental approaches are being developed that will hopefully pro- ing room impulse responses with a high spatial resolution. In vide additional insight into the complexity of such sound field. this paper, the development of such a measurement approach This paper presents the development of a scanning measure- is presented. Several challenges and technical details are prement approach to study the reverberant sound field formed in sented in connection with the approach, including the method the regular chamber that is part of the InnoRenew CoE Acous- for accurately determining the coordinates of the moving microtics Laboratory. The paper is organized to present the theoretical phones. The developed experimental technique is used to study background of the reverent sound fields in Section 2, the regu- the characterization of the sound field. The study is therefore lar chamber facility in Section 3, the developed scanning micro-important to better understand concepts related to the diffuse phone system in Section 4, and the conclusions in Section 5. and reverberant sound field, such as diffuseness and mixing, in the context of modal and statistical descriptions. 2. INVESTIGATING THE REVERENT SOUND FIELD Keywords: Acoustic scanning, measurement automatisation, microphone cable robot Several diffuseness quantifiers have been proposed in the liter- ature, focusing on different sound field properties: 1. INTRODUCTION • spatial correlation [4–7] The diffuse sound field is an important idealized field in acoustics • response decay fluctuations [8–11] for standardized acoustic measurements, simple predictions, • intensity distribution [12, 13] and a wide range of theoretical derivations. Surprisingly, there • directional characteristics [14–17] is no general agreement on the definition of the diffuse sound field, its quantification and characterization; also, there is lim- • spatial uniformity of sound pressure [5] ited understanding on the required features of room boundaries • spatial uniformity of reverberation time [18] that would lead to a diffuse sound field. Controversially, de- mands for its presence are given for a broad range of spaces and • and higher order parameters [19–24] facilities. These quantifiers are capable of evaluating certain sound field Misconceptions about the diffuse sound field were exposed features, but none of them is complete enough to evaluate all already by Schultz [1] in the 1970s and are still present, as aspects of the diffuse sound field. pointed out by Jeong [2]. Several uncertainties exist with re- An important branch of research focuses on the character- spect to various aspects, with the largest knowledge deficit aris- ization of the sound field based on its spatial features. Experi- ing from the absence of a predictable and measurable quanti- mental evaluation of these features involves various measure- fier of the sound field diffuseness, i.e., the degree to which the ment methods, ranging from optical methods [25], to various sound field is diffuse. Such a quantifier would allow the eval- sophisticated microphone scanning approaches. In this context, uation of sound field environments that are currently only as- scanning is the process of systematically acquiring the room im- sumingly diffuse, such as reverberation chambers. These are pulse response over a predefined grid of points in the volume of Rok Prislan: Experimental sound field characterization with automated high-resolution impulse response measurements 148 the room. Scanning is typically performed by robots [26,27] that Figure 1. The rotation of the non-fixed bounding surfaces is 10◦, move one or more microphones across the scanned volume of with the axis of rotation at the center of the surface to minimize the room. changes in the volume of the room. A total of 6 different regular To speed up the scanning process, multiple microphones ar- chamber configurations are possible, as shown schematically in ranged in a microphone array can be used simultaneously. The Figure 2. advantage of using multiple microphones is that the measure- ment process is faster by a factor equal to the number of mi- crophones. The measurement time becomes relevant when the conditions in the room may vary, for example, when the air tem- perature or humidity in the room changes during the measure- ment. An important technical requirement for using multiple microphones is that the microphones are phase matched in the frequency range of interest. A disadvantage of using multiple microphones is that the microphone array generally presents a larger disturbance to the sound field, since such a structure can be relatively large. In addition to sophisticated robot-based methods, an accessible method for scanning the sound field was used by Prislan and Svešnek [28] to visualize modal shapes in the room. The acquisition was based on vertically stacked mi- crophones that were manually moved across the floor plan of the room. With post-processing of the large number of acquired im- pulse responses, it is possible to perform various sound field characterizations. The decomposition into plane or spherical waves is an important approach [29] because it allows us to ex- tract sound field quantities, namely velocities and intensities, in addition to sound pressure. This also allows us to study the directional properties of the sound field, such as the isotropy of Fig. 1. A graphical representation of the rotation of the the reverberant field [26]. Another advantage of the large num-ceiling (red) and two walls (green and yellow) of the reg- ber of measurements generated by scanning is the possibility ular chamber. The remaining walls and the floor are fixes to apply statistics on various sound field characterization meth- and perpendicular to each other. The axis of rotation of ods, such as the spatial uniformity methods [17] or the sound the non-fixed surfaces is in the center of each surface and field sensitivity method proposed by Prislan at. el. [19]. marked by a longer line. 3. THE TEST FACILITY Another important aspect to mention is the expected influ- ence of the changes in the geometry of the regular chamber on Experimental studies with the cable robot will be conducted in the sound field. For this purpose, the modal shapes and eigen- the regular chamber, a recently introduced test facility [30], that frequencies have been previously studied using FEM [30]. It has is part of the InnoRenew CoE Acoustic Laboratory. The facility been shown that the distribution of the room resonances and represents a highly controlled acoustic environment whose ge- the corresponding modal shapes do not differ significantly be- ometry can be changed without drastically altering the volume, tween the different chamber configurations. surface area, or other acoustically relevant properties of the chamber. Such geometric variability is rare in ordinary rooms, which are inherently static and whose geometry can be changed 4. THE DEVELOPMENT OF THE SCANNING MICROPHONE only by adding elements to the room. SYSTEM The basic geometry of the regular chamber is a cuboid with 4.1 The use of robots in room acoustic measurements volume V = 37m3 and internal dimensions l = 3m in length, w = 2.5m in width and h = 2m in height. The ratio of the reg- Sound field scanning, i.e., the acquisition of impulse responses ular chamber dimensions corresponds to the Bolt area [31]. The with high spatial resolution, has previously been approached regular chamber consists of six plane surfaces, i.e. 4 walls, the with various types of robots. Witew et. al. [27] installed a frame floor and the ceiling. Two walls and the floor are perpendicular structure in the Eurogress hall in Aachen, Germany. The frame to each other and are fixed, while the remaining two walls and allowed a cart with 32 linearly arranged microphones to move the ceiling can be tiled independently, as shown schematically in in the horizontal plane of 5.30m by 8.00m. Rok Prislan: Experimental sound field characterization with automated high-resolution impulse response measurements 149 Fig. 2. The 6 geometric configurations of the regular chamber (R1-R6). An arm robot moving a single microphone has been widely used for various investigations at the Technical University of Denmark [32], including measurements in a reverberation chamber, in an anechoic chamber, and in the field. When mea- suring with the robotic arm, the movement of the microphone is not limited to a vertical plane, but can follow any predefined spatial configuration. On the other hand, the movement of the robot arm is limited in range due to the limited size of the arm. An important advantage of using robots for sound field scan- ning measurements is the high accuracy in positioning the mi- crophone in the sound field. In contrast, manual methods [28], are, due to inaccurate microphone positioning, limited to low frequencies, where accuracy is always sufficient compared to the wavelength of interest. 4.2 Microphone positioning system Fig. 3. A graphical representation of the impulse response measurements principle used for microphone positioning The accuracy with which microphone coordinates are known [33] using four high frequency loudspeakers at known po- during scanning measurements is important because coordi- nates can be an input to post-processing analysis [28,29]. To mit- sitions in the room. igate positioning accuracy issues, a variety of microphone posi- tioning systems have been used to date. One purely acoustic approach [33] is to use impulse response measurements to de- 4.3 Cable robot termine the distance between the microphone and a set of fixed Cable-driven parallel robots are a type of parallel manipulator loudspeakers at the known location in the room. From this, tri- that uses flexible cables as actuators [35]. The advantages of ca- angulation can be used to determine the coordinates of the mi- ble robots are the relatively large span in which motion can be crophone, as shown schematically in Figure 3. The accuracy of achieved. A less relevant advantage for the intended applica- this method can be further improved by using multiple micro- tion is also the fast movement capability of cable robots. On the phones in a fixed configuration. other hand, cable robots are less accurate in positioning com- An alternative method for microphone positioning that has pared to other types of robots due to the large spans over which already been investigated is the use of a laser scanner [34]. This the cables can considerably elastically extend at high tensions. involves placing colored markers on the microphone array (see It was therefore a design decision to combine the installation of Figure 4), the coordinates of which can be used to determine the the cable robot with the previously developed acoustic position- coordinates of the individual microphones in the array. ing system presented in Section 4.2. Rok Prislan: Experimental sound field characterization with automated high-resolution impulse response measurements 150 8 cables. In this context, the length of the cables is considered as the distance between the four attachment points of the ca- bles on the array (T1, T2, T3 and T4) and the eight fixed at- tachment points at the room boundaries (P1, P2, ...P8). The lengths can be calculated as linear norms of the vector corre- sponding to the ends of each cable. A specific feature of the introduced design is that two cables are attached to the same fixation point on the microphone array. Currently, the array is assumed to move in parallel to one of the fixed walls and the floor of the chamber. This is an added constrain to the motion of the cable robot, which enables that the position of each microphone in the array is fully determined from the coordinate of the center of the array T. As such, an additional computation step is needed to determine T1, T2, T3 and T4 from T based on the dimensions of the array and its predetermined orientation in space. Another aspect to consider is the positioning of the attach- ment points P1, P2, ...P8 on the boundaries, which represent the outer limit of the motion range of the microphone array. To enable an effectively scan a cuboidal volume, the eight attach- ment points have been put in the proximity of the corners of the chamber. Fig. 4. A laser scan of the back side of the microphone ar-5. CONCLUSIONS ray with visible colored markers, from which microphone This paper presents the motivation for developing an automated coordinates can be extracted using simple geometrical re- sound field scanning system, i.e., a system for measuring the lations [34]. impulse response in a predefined volume with high spatial res- olution. Previous approaches using robots for sound field scan- ning have been described in the context of the robotisation type Different cable robot configurations are possible, differing and post-processing analysis, which is driving the development in the number of cables used and the distribution of fixation of such advanced acoustic instrumentation. points at the room boundaries. In the case of the designed ca- The limitations of accurate positioning of microphones with ble robot, 8 cables are used to control the microphone array, as a cable robot were discussed and the acoustic positioning sys- shown graphically in Figure 5. This motion system is overdeter- tem was presented as an additional system to the cable robot. mined, as the top 4 cables would be sufficient to fully control the The aspect of accurate positioning was also considered in the position of the microphone array. The additional bottom cables design by the introduction of redundant cables, which are ex- were added to stabilize the position of the system, prevent ax- pected to stabilize the positioning of the microphones and also ial rotation of the microphone array, and reduce the oscillatory limit the unwanted rotations and oscillations of the microphone motion that would likely occur with a pendulum-type design of array. the system. The developed model of the cable robot installation In further steps, the developed cable robot will be installed is shown graphically in Figure 5. in the regular chamber so that the sound field scanning can be At the current state of the development, the microphone ar- performed in a highly controlled reverberant environment. In ray to be used still needs to be determined. In fact, an important addition, the acoustic microphone positioning system will be in-aspect is the integration of the cables, which should not inter- tegrated to autonomously determine the microphone coordi- fere with the movement of the cable robot and at the same time nates with high accuracy. Furthermore, the microphone array allow a reliable connection of the microphones to the data ac- will be further developed, paying special attention to the cable quisition system. At the same time, it is important to design the management system, which must flexibly support the required array as a miniature structure that minimally disturbs the sound range of motion of the cable robot. With the developed mea- field under investigation. surement system, we intend to perform various investigations Including the geometrical considerations into the discus- in the reverberant sound field, hoping to gain additional insight sion, it is apparent that the desired coordinate of the center of needed to better understand the concept of the diffuse sound the microphone array T is fully determined by the length of the field. Rok Prislan: Experimental sound field characterization with automated high-resolution impulse response measurements 151 7. REFERENCES [1] T. Schultz, “Diffusion in reverberation rooms,” Journal of Sound and Vibration, vol. 16, pp. 17–28, 5 1971. [2] C.-H. Jeong, “Diffuse sound field: challenges and misconcep- tions,” pp. Pages 1015–1021, 2016. [3] M. Vercammen, “On the revision of iso 354, measurement of the sound absorption in the reverberation room,” Deutsche Gesellschaft für Akustik, 2019. Zweitveröffentlicht auf dem Publikationsserver der RWTH Aachen University. - Weitere Konferenz: 4th EAA Euroregio, 2019-09-09 - 2019-09-13, Aachen, Germany. [4] F. Jacobsen and T. Roisin, “The coherence of reverberant sound fields,” The Journal of the Acoustical Society of Amer- ica, vol. 108, pp. 204–210, 2000. [5] H. Nélisse and J. Nicolas, “Characterization of a diffuse field in a reverberant room,” The Journal of the Acoustical Society of America, vol. 101, pp. 3517–3524, 6 1997. [6] K. Bodlund, “A new quantity for comparative measurements concerning the diffusion of stationary sound fields,” Journal of Sound and Vibration, vol. 44, pp. 191–207, 1976. [7] R. K. Cook, R. V. Waterhouse, R. D. Berendt, S. Edelman, and M. C. Thompson, “Measurement of correlation coefficients in reverberant sound fields,” The Journal of the Acoustical Society of America, vol. 27, pp. 1072–1077, 1955. [8] C.-H. H. Jeong, J. Brunskog, and F. Jacobsen, “Room acous- tic transition time based on reflection overlap,” The Journal of the Acoustical Society of America, vol. 133, pp. 015002– 015002, 5 2013. Fig. 5. A 3D model of the developed cable robot consisting [9] T. Hanyu, “Analysis method for estimating diffuseness of of 8 cables (top: distant view, bottom: zoomed view). The sound fields by using decay-cancelled impulse response,” cables are attached in pairs to the four attachment points Building Acoustics, vol. 21, pp. 125–133, 2014. of the microphone array, forming a rectangular shape. [10] K. Eda and T. Sakuma, “A theoretical survey on temporal The position of the microphone array is fully determined structure of reflections in decay-cancelled room impulse re- by the lengths of the 8 cables. sponses,” Acoustical Science and Technology, vol. 36, p. 453, 2015. [11] C.-H. Jeong, F. Jacobsen, and J. Brunskog, “Thresholds for the slope ratio in determining transition time and quantifying 6. ACKNOWLEDGMENTS diffuser performance in situ,” The Journal of the Acoustical Society of America, vol. 132, pp. 1427–1435, 9 2012. The authors gratefully acknowledges the European Commission [12] G. D. Galdo, M. Taseska, O. Thiergart, J. Ahonen, and for funding the InnoRenew project (grant agreement #739574) V. Pulkki, “The diffuse sound field in energetic analysis.,” under the Horizon2020 Widespread-Teaming program and the The Journal of the Acoustical Society of America, vol. 131, Republic of Slovenia (investment funding from the Republic of pp. 2141–2151, 2012. Slovenia and the European Union from the European Regional Development Fund). The authors acknowledge the financial [13] M. Nolan, E. F. Grande, and C.-H. Jeong, “Characterization of support from the Slovenian Research Agency (research core diffusivity based on spherical array processing,” 2015. funding No. Z1-4388, Toward better understanding the diffuse [14] B. N. Gover, J. G. Ryan, and M. R. Stinson, “Measurements of sound field, and research core funding No. J4-3087, Engineered directional properties of reverberant sound fields in rooms wood composites with enhanced impact sound insulation per- using a spherical microphone array,” The Journal of the formance to improve human well being). Acoustical Society of America, vol. 116, pp. 2138–2148, 2004. Rok Prislan: Experimental sound field characterization with automated high-resolution impulse response measurements 152 [15] A. Politis, S. Delikaris-Manias, and V. Pulkki, “Direction-of- [29] E. G. Williams and J. A. Mann, “Fourier acoustics: Sound ra- arrival and diffuseness estimation above spatial aliasing for diation and nearfield acoustical holography,” The Journal of symmetrical directional microphone arrays,” pp. 6–10, 4 the Acoustical Society of America, vol. 108, 2000. 2015. [30] R. Prislan and J. Brunskog, “A regular chamber - design of [16] N. Epain and C. T. Jin, “Spherical harmonic signal covariance a test facility for the investigation of the reverberant sound and sound field diffuseness,” 2016. field,” 2023. [17] M. Nolan, E. Fernandez-Grande, J. Brunskog, and C.-H. [31] R. H. Bolt, “Note on normal frequency statistics for rectangu-Jeong, “A wavenumber approach to quantifying the isotropy lar rooms,” The Journal of the Acoustical Society of America, of the sound field in reverberant spaces a),” The Journal of vol. 18, pp. 130–133, 7 1946. the Acoustical Society of America, vol. 143, pp. 2514–2526, [32] E. Fernandez-Grande, S. A. Verburg, and X. Karakonstan- 2018. tis, “Sound field reconstruction: Towards large-scale spatial [18] M. Lautenbach and M. Vercammen, “Can we use the stan- sensing,” 2023. dard deviation of the reverberation time to describe diffu- [33] R. Prislan, “Microphone positioning using acoustic location,” sion in a reverberation chamber?,” The Journal of the Acous- 2020. tical Society of America, vol. 133, p. 3348, 2013. [34] R. Prislan and U. Kavka, “The potential of using a laser [19] R. Prislan, J. Brunskog, F. Jacobsen, and C.-H. Jeong, “An ob-scanner to facilitate in-situ room acoustic measurements,” jective measure for the sensitivity of room impulse response pp. 510–515, 10 2022. and its link to a diffuse sound field,” The Journal of the Acous- [35] A. Pott, Cable-driven parallel robots: Theory and application, tical Society of America, vol. 136, pp. 1654–1665, 2014. vol. 120. 2018. [20] J.-D. Polack, “Modifying chambers to play billiards: the foundations of reverberation theory,” Acta Acustica united with Acustica, vol. 76, 1992. [21] A. Bidondo, N. Xiang, and J. Herder, “Experimental inves- tigation on varied degrees of sound field diffuseness in enclosed spaces,” Proceedings of Meetings on Acoustics, vol. 28, p. 15020, 9 2016. doi: 10.1121/2.0000436. [22] G. Defrance and J.-D. Polack, “Estimating the mixing time of concert halls using the extensible fourier transform,” Ap- plied Acoustics, vol. 71, pp. 777–792, 9 2010. [23] C.-H. Jeong, “Kurtosis of room impulse responses as a dif- fuseness measure for reverberation chambers.,” The Journal of the Acoustical Society of America, vol. 139, p. 2833, 2016. [24] H. Dujourdy and T. P. Jean-Dominique Pialot, Baptiste Toule- monde, “Modified wave equation for modelling diffuse sound field,” 2016. [25] S. A. Verburg and E. Fernandez-Grande, “Acousto-optical volumetric sensing of acoustic fields,” Physical Review Ap- plied, vol. 16, 2021. [26] M. M. Nolan, S. A. Verburg, J. Brunskog, E. F. Grande, and E. Fernandez-Grande, “Experimental characterization of the sound field in a reverberation room,” Acoustical Society of America. Journal, vol. 145, pp. 2237–2246, 4 2019. [27] I. B. Witew, M. Vorländer, and N. Xiang, “Sampling the sound field in auditoria using large natural-scale array measure- ments,” The Journal of the Acoustical Society of America, vol. 141, pp. EL300–EL306, 3 2017. [28] R. Prislan and D. Svenšek, “Room mode shape visualization using a multi-microphone measurement technique,” 2016. Rok Prislan: Experimental sound field characterization with automated high-resolution impulse response measurements 153 Contributed papers Acoustic software and training 1. Audio exercises: quality, pitch statistics and long-term spectra of the sound files Andrea Andrijasevic (Polytechnic of Rijeka, Croatia) 2. Acoustics Knowledge Alliance project: the development of open-access interactive online educational materials in acoustics - strategy and results Marko Horvat (University of Zagreb, Faculty of electrical engineering and computing) 154 AUDIO EXERCISES: QUALITY, PITCH STATISTICS AND LONG-TERM SPECTRA OF THE SOUND FILES Andrea Andrijašević1, Mirta Čulina2 1 Polytechnic of Rijeka, Vukovarska 58, 51 000 Rijeka, Croatia 2 Clinical Hospital Center Rijeka, Krešimirova 42, 51 000 Rijeka, Croatia Abstract: Audio exercises is a free computer program created with the objective of supporting hearing therapy for adult persons with hearing impairments and tinnitus. It is intended for their home use and, as such, comprised of two parts: a set of twelve auditory exercises in the Croatian language to be used during the later stages of a person’s adaptation to the prescribed hearing aid or cochlear implant, and a module for tinnitus perceived intensity reduction based on the Tailor-made notched music training (TMNMT) sound therapy approach. In addition to the program’s main task of helping in patient’s rehabilitation, it can also be used by the non-diagnosed individuals as a relatively straightforward monitoring tool that can inform them early of their hearing health deterioration. Given these manifold valuable potential applications of the program, in this paper we provide an analysis of the sound files quality as well as of the pitch statistics and long-term spectra of the speech and music examples provided in it. Since the auditory exercises are based on the verbotonal method that takes into account the optimal hearing frequencies of speech sounds, we also calculate the log-spectral distance between the long-term average spectra of the five classes of speech examples and, similarly, the distances between the spectra of the five music genres used for tinnitus therapy. Keywords: hearing loss, tinnitus, rehabilitation, music therapy, computer program 1. INTRODUCTION therapy principles [6]. As shown in Figure 1, the main program window contains three items: According to the projections presented in the World ● Vježbe slušanja (auditory exercises), which Health Organisation’s latest report on hearing, by 2025 consists of a set of twelve auditory exercises nearly a quarter of the world population will have some ( Fraze - Veliki test 2), degree of hearing loss [1]. Moreover, it is anticipated that ● Terapija tinitusa (tinnitus therapy), one in fourteen people will require hearing care [1]. ● Bilješke (notes). Hearing loss and tinnitus, the perception of a sound without an external source, besides the negative impact they have on speech intelligibility, commonly have a negative effect on both the psychological state and social life of affected individuals [2, 3]. Since for a part of the affected population the medical support, therapy and counselling can be expensive and time-consuming [4], a computer program titled “Audio exercises” was recently created with the intention of providing free and accessible support for the Croatian speaking adults [5]. It was written in C# and runs on Windows 7 and 10 operating systems. The program contains speech material for hearing exercises as well as pieces of instrumental music filtered according to the Tailor-Made Notched Music Training (TMNMT) tinnitus Fig. 1. Main program window Andrijašević et al.: Audio exercises: quality, pitch statistics and long-term spectra of the sound files 155 AAAA – 2023 – IZOLA - Conference Proceedings In addition to the program’s main task of helping in subtracted. In this way, two SNRs were obtained for a patients rehabilitation, it can also be used by the non- speech file - one using the noise power estimate from the diagnosed individuals as a relatively straightforward file beginning (SNR-B) and the second one using the file monitoring tool that can inform them early of their end noise power estimate (SNR-E). Figure 2 shows the hearing health deterioration. Given these valuable SNRs and their exercise-level mean. potential applications of the program, in this paper we provide an analysis of the phonetical and spectral Phoneme Frequency range Croatian language characteristics of its sound files. class (Hz) phonemes The remainder of the paper is organised as follows. In Section 2, we first present the phonetic content, sound 150 - 300 /m/, /n/, /nj/ quality, and spectral characteristics of the recorded low (L) 200 - 400 /b/, /p/, /u/ speech material used in the program. We continue by analysing the spectral characteristics of the pieces of 300 - 600 /v/ low-mid instrumental music used for tinnitus therapy in Section 3. 400 - 800 /g/, /o/ (LM) Finally, with Section 4, we conclude our work. 600 - 1200 /h/, /l/, /lj/ 800 - 1600 /a/, /k/, /r/ mid (M) 2. SPEECH MATERIAL 1200 - 2400 /d/, /dž/, /f/, /m/, /ž/ mid-high 1600 - 3200 /č/, /e/, /n/, /lj/, /š/, /t/ The set of twelve auditory exercises is intended for patient (MH) 2400 - 4800 /đ/, /j/, /nj/ use at home during the later stages of their adaptation to the prescribed hearing aid or cochlear implant. The 3200 - 6400 /ć/, /i/ exercises are based on the verbotonal hearing therapy high (H) 4800 - 9600 /c/, /z/ method that takes into account the patient’s optimal 6400 - 12800 /s/ frequency range of hearing [7, 8]. Accordingly, the speech material used in them was carefully selected in order for it to contain an equal number of speech sounds from all Table 1. Frequency range-based phoneme classes frequency classes. It is in Croatian, and was read by a female person with an average sound pressure level of 71 dB in an anechoic chamber, recorded using the LingWaves system with a sampling rate of 22050 Hz. For the analyses presented in Sections 2.3. - 2.5., the words and sentences used in exercises 6 - 12 needed to be separated into five classes based on the frequency range in which most of their energy is contained. This was performed in line with the classification of the Croatian language phonemes given in [9], an example of which is shown in Table 1. 2.1. Sound quality The sound quality of speech files was evaluated using the estimates of their signal-to-noise ratio (SNR). To this end, Fig. 2. SNR-B and SNR-E of the sound files. Filled symbol noise power was estimated from both the first and last 50 denotes the corresponding exercise-level mean ms of a file (i.e., from the parts with no voice activity), whereas speech signal power was estimated as the active In the speech enhancement community, the general speech level using the ITU-T P.56 Method B [10, 11] from consensus is that speech signals with an SNR greater than which the previously estimated noise power was 30 dB can be considered “clean”, meaning they do not Andrijašević et al.: Audio exercises: quality, pitch statistics and long-term spectra of the speech files 156 AAAA – 2023 – IZOLA - Conference Proceedings require additional processing since above that value noise phonemes. A higher level of similarity between the and reverberation do not reduce speech intelligibility or phonetic content distributions can be observed in Figure degrade the performance of systems for automatic 5, where around 50 % of the phonemes belong to the class speech recognition. This threshold was set even lower in the sentence had previously been labelled with. Finally, two recent speech signal processing challenges [12, 13], for the recordings of words, the distributions presented in where signals with an SNR of 20 dB were provided as the Figure 6 show larger variation due to a smaller total best case scenario. It can therefore be concluded that the number of speech sounds present in the recordings, but, recorded speech material is of excellent quality since 98 % at the same time, it should be noted that the percentage of the files have SNR higher than 30 dB. of the phonemes that belong to the word label class is, on average, higher than in the former figure. 2.2. Pitch statistics Pitch, i.e. the fundamental frequency (F0) curves of speech files were estimated using the Voicebox function v_fxpefac [11]. Figure 3 presents their mean, standard deviation and median F0, as well as the exercise-level mean. Overall, these pitch statistics indicate that there was no significant change in neither mean F0 nor its variation across the exercises. The grand mean F0 is slightly under 210 Hz with a standard deviation of 36 Hz. Fig. 3. Mean, standard deviation and median of the F0 curves. Filled symbol denotes the corresponding exercise-level mean 2.3. Phonetic content Figures 4, 5, and 6 present the distributions of phonetic content of the speech files, based on the phoneme classes Fig. 4. Distributions of phonetic content of the speech defined in Table 1. In the upper part of Figure 4, the files of the first five auditory exercises. Due to Matlab distributions are quite similar across the files since all of limitations, the following notation was used for a them contain text “vidimo se u”, whereas the other part phoneme subset: /lj/ as /L/, /nj/ as /N/, /č/ as /O/, /ć/ as of speech material consists mostly of mid and mid-high /C/, /ž/ as /Z/, /š/ as /S/, /dž/ as /D/, and /đ/ as /X/ Andrijašević et al.: Audio exercises: quality, pitch statistics and long-term spectra of the speech files 157 AAAA – 2023 – IZOLA - Conference Proceedings Fig. 5. Distributions of phonetic content of sentences Fig. 6. Distributions of phonetic content of words saturated with phonemes of one class saturated with phonemes of one class Andrijašević et al.: Audio exercises: quality, pitch statistics and long-term spectra of the speech files 158 AAAA – 2023 – IZOLA - Conference Proceedings Figures 7, 8, and 9 show the phonetic content dissimilarity (PCD) matrices for pairs of speech recordings, where the dissimilarity is calculated as the ratio of the number of phonemes that the longer utterance (i.e., the one with the higher speech sounds count) does not share with the shorter utterance and the total number of speech sounds present in the longer utterance. In case the utterances are of equal length, we calculate the PCD based on the second utterance’s data. To illustrate the suitability of this dissimilarity measure, we calculate its value for the pair of words “jedan” and “nedjelja” - the latter, longer word contains 7 speech sounds, of which one /e/ is not shared with the first word, as well as the phoneme /lj/, resulting in a PCD of 0.286, as shown in Figure 7. In the same Figure, the phonetic Fig. 8. Phonetic content dissimilarity matrix of sentences overlap that occurs for the recordings containing text saturated with phonemes of one class “vidimo se u” is clearly visible (dissimilarity being low). Also interesting to notice are the results for the word “hvala”, a utterance very dissimilar in its phonetic content to other utterances present in the first five auditory exercises, the only exception to this being the utterance “dva”, as confirmed with the proposed measure. The trends observed in Figures 5 and 6 can now be, even more easily, discerned in Figures 8 and 9. The phonetic content dissimilarity of the utterances labelled with the L class and other classes increases from left to right (and bottom to top) in both matrices. Even though these two matrices share a similar pattern, the overall PCD is higher for the latter matrix as words contain a smaller number of speech sounds than sentences, which consequently reduces the probability of phoneme overlap. Fig. 9. Phonetic content dissimilarity matrix of words saturated with phonemes of one class Fig. 7. Phonetic content dissimilarity matrix of the speech files of the first five auditory exercises Andrijašević et al.: Audio exercises: quality, pitch statistics and long-term spectra of the speech files 159 AAAA – 2023 – IZOLA - Conference Proceedings 2.4. Long-term average spectrum Figures 10, 11, and 12 show the long-term average spectrum (LTAS) [14] of the speech files obtained using a sliding 50 ms long Hamming window with 75 % overlap, and smoothed with a 1/20 octave Gaussian window [15]. It can be observed that the LTAS varies across the files depending on their phonetic content. To illustrate this, one can examine the LTAS of two utterances that differ in only one phoneme – “devet” (nine) and “deset” (ten) in Figure 10. The former utterance contains the voiced approximant /v/, whereas the latter one the voiceless fricative /s/ [16, 17, 18]. Due to the manner of its production, fricative /s/ is a sibilant speech sound of high energy [19], so that in the 5 - 10 kHz region the word “deset” contains considerably more energy than the word “devet”. The remaining Croatian sibilants /c/, /z/, /č/, /ž/, and /š/ [19] also generate large LTAS values in the region above 2 Fig. 11. Long-term average spectrum of sentences kHz, for example in the words “šest”, “četiri”, and saturated with phonemes of one class “četvrtak”. Similar LTAS shapes can also be observed in Figures 11 and 12, whenever a utterance contains one or more sibilant speech sounds. Fig. 12. Long-term average spectrum of words saturated with phonemes of one class Fig. 10. Long-term average spectrum of the speech files of the first five auditory exercises Andrijašević et al.: Audio exercises: quality, pitch statistics and long-term spectra of the speech files 160 AAAA – 2023 – IZOLA - Conference Proceedings To describe the differences in LTAS observed in Figures 10, 11, and 12 with a single number, we calculate the log- spectral distance (LSD) between LTAS pairs using the following formula: 1 𝜋 𝐿𝑆𝐷 = √ ∙ ∫ [𝐿𝑇𝐴𝑆 2𝜋 1(𝜔) − 𝐿𝑇𝐴𝑆2(𝜔)]2d𝜔 −𝜋 As can be seen in Figures 13, 14, and 15, this measure successfully captures the differences. Due to the presence of sibilants, the pairs of previously mentioned utterances from Figure 10 have small LSDs in Figure 13. In Figure 14, smaller intra-class and larger inter-class LSD can be observed for the sentences. As expected, for the speech files of the first class (L), the LSD increases from Fig. 14. Log-spectral distance matrix of sentences left to right (and bottom to top). An interesting exception saturated with phonemes of one class to this are the utterances “umoran konj vuče plug” and “božo hoda okolo” whose LTAS is closer to the LTAS of the M and MH class utterances due to presence of the sibilants /č/ and /ž/. Finally, the LSD matrix for the auditory exercises containing words is presented in Figure 15. Again, inter- class LTAS similarity can be noticed, although now with a higher degree of similarity between the utterances labelled with L, and LM, and M classes than observed in Figure 14. Fig. 15. Log-spectral distance matrix of words saturated with phonemes of one class Fig. 13. Log-spectral distance matrix of the speech files of the first five auditory exercises Andrijašević et al.: Audio exercises: quality, pitch statistics and long-term spectra of the speech files 161 AAAA – 2023 – IZOLA - Conference Proceedings 2.5. Relationship between PCD and LSD variation in LSD is quite small in the first cluster, whereas the more diverse spectral profiles observed for the In this section, we explore the relationship between the classical and old time music pieces in Figure 18 are a direct PCD and LSD. To this end, use only the PCD and LSD data result of a more diverse instrumentation utilised in these obtained for the pairs of utterances that contain the same genres than in the former ones. Specifically, the files 29, number of speech sounds. Figures 16 and 17 show the 31, 34, 36, 39, 40, and 42 - 45, that differ most significantly scatter diagrams in which the number of speech sounds is from the files of the first three genres, are all either piano colour coded. Even though the relationship between PCD or guitar solo pieces, whereas the remaining pieces are and LSD for utterances with five or less speech sounds is performed by ensembles of different instruments, and blurry, it becomes more clear as the number of speech therefore more similar to pieces of the first cluster. sounds increases. Observed blurriness can be attributed primarily to a lack of manner of sound production class encoding in the PCD measure. For example, even though the phonemes /t/, /e/, and /š/ belong to the same class (MH), their spectral profiles differ due to the manner of their production [17]. The effect of this lack of encoding slowly decreases as the number of speech sounds in a utterance increases. 3. INSTRUMENTAL MUSIC The tinnitus therapy part of the program contains instrumental music pieces of five genres - pop, rock, electronic, classical, and old time, downloaded from the freemusicarchive.org web page. In accordance with the Fig. 16. Relationship between PCD and LSD for pairs of TMNMT therapy principles, music files were filtered in utterances consisting of five or less speech sounds. The Praat [20] using octave wide notch filters. In order to number of speech sounds is colour coded reduce the program complexity while retaining a perceptually appropriate resolution of tonal tinnitus frequencies [21, 22], the notch filter centre frequency corresponds to one of the ISO 266 standard one-third octave band centre frequencies in the 2 - 10 kHz range [23]. The user first identifies the frequency that most closely matches their tonal tinnitus frequency among the aforementioned centre frequencies, after which they can download the suitably filtered music files. 3.1. Long-term average spectrum In Figure 18 we present the LTAS of the music files obtained using a sliding 50 ms long Hamming window with 50 % overlap. It can be noticed that the spectral profiles differ considerably. To quantify these differences, in Figure 19 we present the corresponding LSD matrix. Two Fig. 17. Relationship between PCD and LSD for pairs of clusters can be detected - the first one for the pop, rock utterances consisting of six or more speech sounds. The and electronic music pieces, and the second one for the number of speech sounds is colour coded pieces of the remaining two genres. Interestingly, the Andrijašević et al.: Audio exercises: quality, pitch statistics and long-term spectra of the speech files 162 AAAA – 2023 – IZOLA - Conference Proceedings 5. REFERENCES [1.] The World Report on Hearing. Available at: https://www.who.int/publications/i/item/97892400 20481 (Accessed 28 August 2023). [2.] Reis, M., McMahon, C. M., Távora-Vieira, D., Humburg, P., Boisvert, I. Effectiveness of Computer- Based Auditory Training for Adult Cochlear Implant Users: A Randomized Crossover Study, Trends in Hearing, 25, 2021. [3.] Bonetti, L., Ratkovski, I., Šimunjak, B. Suvremena rehabilitacija odraslih osoba sa stečenim oštećenjem sluha, Liječ Vjesn, 139, pp. 292-298, 2017. Fig. 18. Long-term average spectrum of the music files [4.] Tuz, D., Isikhan, S. Y. and Yücel, E. Developing the computer-based auditory training program for adults with hearing impairment, Med Biol Eng Comput, 59, pp. 175–186, 2021. [5.] Čulina, M. and Andrijašević, A. Audiovježbe: računalni program za vježbe slušanja i terapiju tinitusa, Zbornik Veleučilišta u Rijeci, 11 (1), pp. 331-351, 2023. Available at: https://doi.org/10.31784/zvr.11.1.18 [6.] Okamoto, H., Stracke H., Stoll W. and Pantev C. Listening to tailor-made notched music reduces tinnitus loudness and tinnitus-related auditory cortex activity, Proceedings of the National Academy of Sciences, 107(3), pp. 1207–1210, 2010. [7.] Guberina, P. Govor i čovjek: Verbotonalni sistem. Poliklinika za rehabilitaciju slušanja i govora SUVAG, ArTresor naklada, 2010. [8.] Guberina, P. Verbo-tonalna metoda u audiologiji. Zavod za fonetiku, 1965. [9.] Rulenkova, L. I. Kako malo gluho dijete naučiti slušati Fig. 19. Log-spectral distance matrix of the music files i govoriti primjenom verbotonalne metode. Poliklinika SUVAG, 2015. [10.] Objective Measurement of Active Speech Level, 4. CONCLUSION International Telecommunications Union (ITU-T), Recommendation P.56, 1993. In this paper we have presented a short overview of the [11.] Brookes, D. M. VOICEBOX: A speech processing ”Audio exercises”, a computer program for hearing and toolbox for MATLAB, Available at: tinnitus therapy. We have shown that the speech http://www.ee.ic.ac.uk/hp/staff/dmb/voicebox/voic recordings used for hearing therapy are of satisfactory ebox.html quality, both SNR- and pitch statistics-wise. Furthermore, [12.] Eaton, J., Gaubitch, N. D., Moore, A. H. and Naylor, P. we also have highlighted the importance of assessing the A. Estimation of room acoustic parameters: The ACE (dis)similarity of recordings from multiple perspectives - Challenge, IEEE/ACM Trans. Audio, Speech, and one of them being phonetical, using the newly proposed Language Process., 24(10), pp. 1681-1693, 2016. measure of phonetic content dissimilarity, and the other [13.] Kinoshita, K., Delcroix, M., Gannot, S., Habets, E. A. one spectral, using the log-spectral distance of their long- P., Haeb-Umbach, R., Kellermann, W., Leutnant, V., term average spectra, as they provide complementary Mass, R., Nakatani, T., Raj, B., Sehr, A. and Yoshioka, insights. Finally, even though music genre is not a deciding T. A summary of the REVERB challenge: state-of-the- factor, the spectra of the instrumental music pieces do art and remaining challenges in reverberant speech indicate that it should perhaps be advised to the tinnitus processing research, EURASIP Journal on Advances in patients to listen to the music pieces performed by Signal Processing, 7, pp. 1-19, 2016. ensembles rather than solo pieces as the former provide [14.] Rose, P. Forensic Speaker Identification. Taylor & richer spectral content. Francis, 2002. Andrijašević et al.: Audio exercises: quality, pitch statistics and long-term spectra of the speech files 163 AAAA – 2023 – IZOLA - Conference Proceedings [15.] Institute of Sound Recording, University of Surrey. [19.] Jelaska, Z. Fonološki opisi hrvatskoga jezika. IoSR Matlab Toolbox. Available at: Hrvatska sveučilišna naklada, 2004. https://github.com/IoSR-Surrey/MatlabToolbox [20.] Boersma, P. and Weenink, D. Praat: doing phonetics [16.] Horga, D. Hrvatsko i slovensko /v/ u akustičkoj by computer, version 6.1.55. University of usporedbi, in Slovenski javni govor in jezikovno- Amsterdam, 2021. Available at: kulturna (samo)zavest, Obdobja 38. Ljubljana: https://www.fon.hum.uva.nl/praat/ Znanstvena založba Filozofske fakultete, pp. 27-37, [21.] Ueberfuhr, M. A., Wiegrebe, L., Krause, E., Gürkov, R. 2019. and Drexl, M. Tinnitus in Normal-Hearing [17.] Quatieri, T. F. Production and classification of Participants after Exposure to Intense Low- speech sounds. In Discrete-Time Speech Signal Frequency Sound and in Ménière's Disease Patients, Processing: Principles and Practice. Prentice Hall, pp. Front Neurol., 7(239), pp. 1-11, 2017. 55-110, 2001. [22.] Pinel, J. P. J. Biološka psihologija. Naklada Slap, [18.] Andrijašević, A. Effect of phoneme variations on 2002. blind reverberation time estimation, Acta Acustica, [23.] ISO 266:1997. Acoustics - Preferred frequencies, 4(1), pp. 1-17, 2020. Available at: https://acta- International Organization for Standardization, 1997. acustica.edpsciences.org/articles/aacus/abs/2020/0 1/aacus200001s/aacus200001s.html Andrijašević et al.: Audio exercises: quality, pitch statistics and long-term spectra of the speech files 164 ACOUSTICS KNOWLEDGE ALLIANCE PROJECT: THE DEVELOPMENT OF OPEN-ACCESS INTERACTIVE ONLINE EDUCATIONAL MATERIALS IN ACOUSTICS - STRATEGY AND RESULTS Marko Horvat1, Lukas Aspöck2, Andreas Herweg3, Kristian Jambrošić1, Karolina Jaruszewska4, Manuel Melon5, Antonio Petošić1, Yannick Sluyts6, Thomas Wulfrank7, Seweryn Zeman8 1University of Zagreb Faculty of EE and Computing, Unska 3, 10000 Zagreb, Croatia 2Institute for Hearing Technology and Acoustics, RWTH Aachen, Kopernikusstraße 5, 52074 Aachen, Germany 3HEAD acoustics GmbH, Ebertstraße 30a, 52134 Herzogenrath, Germany 4KFB Acoustics Sp. z o.o., Mydlana 7, 51502 Wroclaw, Poland 5Laboratoire d’Acoustique de l’Université du Mans, UMR CNRS 6613, Avenue Olivier Messiaen, 72000 Le Mans, France 6KU Leuven, Department of Architecture, Campus Brussels and Ghent, Paleizenstraat 65/67, 1030 Brussels, Belgium 7Kahle Acoustics, Avenue Molière 188, 1050 Bruxelles, Belgium 8Jazzy Innovations Sp. z o.o., ul. Zygmunta Starego 22/42, 44100 Gliwice, Poland Abstract: In 2020, an interdisciplinary consortium made up of four universities and four companies started the Acoustics Knowledge Alliance (ASKNOW) project funded by the Erasmus+ Programme of the European Union, with the goal to develop freely accessible, online, interactive educational materials in five fields of acoustics, to be accessible and available on the already established Acoustic Courseware (ACOUCOU) online platform. The mission and the vision of the consortium was to bring this knowledge closer to educators, students, professionals in acoustics, but also specialists in other fields, as well as lay people. In particular, the materials were developed in the fields of acoustic fundamentals, psychoacoustics, acoustic simulations and auralization, electroacoustics, and room and building acoustics, to be grouped in five corresponding online courses. The building units that form the developed materials are lessons and practical cases with a theoretical part that gives the underlying theory, a principle part that illustrates the phenomenon being presented, and a task part that contains one or more tasks to be solved. The process of developing educational materials that will reach out to users and present the knowledge in an interesting and captivating way is complex and multidimensional, requiring an interdisciplinary approach. The intent of this paper is to present and discuss the key aspects of the development process. Moreover, as the project has reached its end, the paper will also present the overview of the content that has been developed, with selected examples of the developed materials in their final, ready-to-use form. Keywords: acoustics, online learning, open courseware, interactive material 1. INTRODUCTION educational institutions have considerable difficulties to keep up with these changes. They base their curricula on In modern times, technology and knowledge change and providing a knowledge basis to build on, but often neglect grow faster and faster, thus imposing the need of life-long the problem-solving approach to gaining knowledge. and fast learning on professionals and specialists in their The COVID 19 pandemic has forced the educators to respective fields. Being fairly inert by nature, the switch to online learning through different means: online Horvat et al.: Acoustics Knowledge Alliance project: the development of open-access interactive online educational materials in acoustics - strategy and results 165 AAAA – 2023 – IZOLA - Conference Proceedings live lectures, pre-recorded material, online platforms, the five courses that cover the corresponding topics. The etc., and the high-speed data transfer enabled this materials are in their final form and are ready to be tested transformation. and perfected before being presented to the public. Fast changes in technology are a challenge to many people who struggle to follow these changes. For this reason, many avoid any kind of contact with the so-called STEM 2. THE KEY ASPECTS OF THE DEVELOPMENT PROCESS field (science, technology, engineering, and mathematics). As a part of STEM, acoustics seems to share the fate of 2.1. General aspects other STEM disciplines. A few years ago, a group of enthusiasts decided to tackle As stated in the introduction, the work within the this situation by creating the Acoustic Courseware ASKNOW project was divided between eight partners, as (ACOUCOU) platform [1], through which they would detailed in the affiliation list. Four of them are universities, disseminate the knowledge in selected fields of acoustics and the remining four belong to the business world and to different target groups such as students and their deal with acoustic consultancy and product development. teachers, professionals, but also to lay people who want In addition, the project employed a number of associated to learn about acoustics. The knowledge is presented in partners who collaborated with the project consortium in form of freely accessible online materials developed over all aspects of the development process. the course of several projects: the Architectural Acoustics The work was divided into thirteen work packages with Multibook (ArAc Multibook), the Acoustic Course for the tasks to: prepare guidelines for developing the Engineers (ACE) and the Acoustic Course for Industry educational materials, manage the project and ensure (ACI). More information on the platform itself and on the smooth execution, develop raw material in five different currently available materials is given in [1-4]. topics, compile all raw material and even it out with The Acoustics Knowledge Alliance (ASKNOW) project [5], regard to form, to develop the courseware in its final funded by the European Union’s Erasmus+ Knowledge form, to test the courseware and make appropriate Alliances programme, is the newest addition to the adjustments, to deal with quality assurance and ACOUCOU platform. Eight partners take part in the project evaluation, and to disseminate the results of the project. and form the project consortium, i.e. four higher To ensure a smooth execution of the project, many forms education institutions and four small and medium of communication were established. Since the project was enterprises. The project consortium is supported by running almost entirely during the COVID-19 pandemic, outside partners. The outcome of this project are the the number of live meetings had to be reduced to a bare freely accessible online teaching materials developed to minimum. Instead, fruitful communication was achieved provide fundamental knowledge in five fields of acoustics, using electronic forms of communication, such as the namely Acoustic fundamentals, Psychoacoustics, Acoustic Mattermost portal and numerous online meetings simulations and auralization, Electroacoustics, and Room executed using available audio/video conferencing tools. and building acoustics. The progress of the project and its timely execution were As one of the project goals is to disseminate its existence monitored via an online progress chart in which the and purpose to a wide range of audience, the status of the progress of each work package was recorded, with the project and its progress and results have already been emphasis on the ones that dealt with the development of reported using different means of dissemination, which courseware material in both raw and final form. include the already published journal and conference The workflow designed for the development of the papers [6-11]. educational materials is illustrated in Figure 1. In the In this paper, the aim of the authors is to present the preparation stage, the main task was to prepare working strategy that has been adopted and carried out guidelines and templates that will be used for the over the course of the project to ensure a smooth and development of raw material. Once developed by experts fruitful creative process. Since the project has already from universities and companies, it was collected and ended, the authors would also like to show representative analysed by teams of designers, who unified the examples of the developed online courseware for each of appearance of the material and designed the appropriate Horvat et al.: Acoustics Knowledge Alliance project: the development of open-access interactive online educational materials in acoustics - strategy and results 166 AAAA – 2023 – IZOLA - Conference Proceedings user interfaces to evoke the best user experience. The which they were involved from the very start to the final work was then taken on by programmers who publication of the finished product. Internal reviews were implemented the proposed solutions and adjusted the intensively conducted during the preparation of raw material for successful use by end users. The material was material and its compilation and conversion to its final repeatedly tested and adjusted based on the received form, as well as during and after the testing had been feedback, to yield the final product. done. Apart from internal reviewing, the material was reviewed by external reviewers and tested by members of target groups for which the product is primarily intended. 2.2. The structure of the educational materials As stated above, the educational materials cover five different topics. Therefore, the logical development was to form as many self-sufficient online courses, each of which covers one topic. Due to their modular nature, these courses can be used as a whole, or selected modules / lessons can be utilized to enhance the teaching / learning Fig.1. The workflow designed for the execution of the process. ASKNOW project To achieve modularity, the educational materials are divided into lessons, each of which covers a single Special attention was paid to the quality of the developed elementary particle of knowledge. Each of the five topics material, which went through numerous stages of testing is covered with thirty such lessons. On top of this, each and reviewing, as illustrated in Figure 2. course contains two practical cases that show common issues, and present ways of solving these issues that stem from the knowledge covered by the lessons. Each lesson consists of three parts. The theory part (part A) uses textual information complemented with equations, figures, and tables to explain the phenomenon covered by the lesson. The principle part (part B) presents the phenomenon through different kinds of audio-visual learning content such as charts, animations, interactive calculations, videos, sound samples one can listen to, etc., or a combination of them. The task part (part C) encourages the user to solve one or more short tasks, which can assume different forms, such as questions with multiple-choice answers, calculation tasks, grouping of images or other objects into appropriate categories, listening to sounds and deciding how they fit certain criteria, etc. The layout of the lessons that has been devised and implemented in the online materials is shown in Figure 3. The main intent is to have the theory part displayed side Fig.2. The flowchart of the testing and reviewing process by side with the principle part (and the task part), so that implemented during the ASKNOW project the user has access to both while studying the phenomenon covered by the lesson. The theory part The partners involved in developing the educational consists of scrollable text the user can browse through. materials had a crucial role in the reviewing process, in The principle part consists of one or more states that are Horvat et al.: Acoustics Knowledge Alliance project: the development of open-access interactive online educational materials in acoustics - strategy and results 167 AAAA – 2023 – IZOLA - Conference Proceedings interlinked and can be called with navigation buttons or 13 - Spherical wave and pulsating sphere; 14 - Monopole, by changing the values of parameters in a way that is the dipole, and quadrupoles; 15 - Solutions of wave equation most appropriate for the content being shown. For in spherical coordinates; 16 - Radiation of a plane surface; example, if a phenomenon is illustrated with four 17 - Diffraction. parameters, each of which can have two values, this yields The third section presents the behaviour of waveguides a total of 16 different states. Each state is automatically and cavities through the following lessons: 18 - Boundary called simply by changing the value of a single parameter. conditions and degrees of freedom; 19 – Waveguides; 20 - Modes in a cavity; 21 - Resonance in a cavity. The final section collects miscellaneous topics that help the user grasp basic concepts of acoustics. For this, the following lessons were developed: 22 - The dB scale; 23 - Sources summation, 24 - Image sources, 25 - Dispersion, group and phase velocity; 26 – Dissipation; 27 - Surface impedance of materials; 28 - Transmission lines and analogies; 29 - Matricial formalism for waveguides; 30 - Horns. 3.2. Psychoacoustics Fig.3. The layout of a lesson adopted for the developed This course is focused on the underlying phenomena educational materials related to human perception of sound and tries to illustrate the applications derived from them. The first section presents the human hearing system via 3. AN OVERVIEW OF THE DEVELOPED CONTENT these lessons: 1 - Anatomy of the ear; 2 - Signal processing of sound by humans; 3 - Bark bands (critical bands); 4 - This section provides a short overview of the content that Masking effects; 5 - Hearing threshold (How it is measured has been covered in the five selected topics and / age-related changes). developed for the five corresponding courses. The courses The second section deals with psychoacoustic parameters are divided into logical subsections containing one or through the following lessons: 6 - Introduction to more lessons. psychoacoustics; 7 - Loudness; 8 - Roughness and fluctuation strength; 9 - Tonality; 10 - Sharpness. 3.1. Acoustic fundamentals The third section is oriented on binaural hearing and measurements. It consists of these six lessons: 11 - This course presents the fundamental concepts in Introduction to binaural hearing; 12 - Sound source acoustics as the theoretical basis for other fields of localization; 13 - Head-related transfer function; 14 - acoustics. Hearing models; 15 - Binaural recording and The first section illustrates the fundamentals of sound and measurement; 16 - Equalization. waves / 1D acoustics (tubes) via these lessons: 1 - The fourth section illustrates the intricacies of speech Definition, what is a wave, speed of sound, physical production and speech intelligibility using these lessons: quantities involved, order of magnitude; 2 - 1D equations 17 - Speech production; 18 - Lombard speech; 19 - Speech of acoustics; 3 - Solutions of acoustic equations in the time in reverberation and noise; 20 - Binaural unmasking; 21 - domain; 4 - Equation of acoustics and their solutions in the Assessment of speech intelligibility (C50, D50, U50, STI). frequency domain – stationary waves, SWR; 5 - Reflection; The fifth section covers the field of psychoacoustic 6 - Transmission and reflection; 7 - Impedance; 8 - Proper listening tests procedures via these five lessons: 22 - frequencies and modes; 9 - Acoustic intensity and losses. Introduction to listening experiments; 23 - Design of The second section deals with 3D waves and sources and listening experiments; 24 - Common experimental consists of the following lessons: 10 - 3D equations of methods (common listening test procedures); 25 - acoustics; 11 - Plane wave in 3D; 12 - Snell-Descartes laws; Horvat et al.: Acoustics Knowledge Alliance project: the development of open-access interactive online educational materials in acoustics - strategy and results 168 AAAA – 2023 – IZOLA - Conference Proceedings Statistical evaluation of listening experiments; 26 - Sound auralisation in research; 28 - Uncertainties of simulations; system quality evaluation testing. 29 - Interactive VR systems; 30 - Perceptual evaluation. The final section strives to explain additional perception- The first practical case displays a common topic in related phenomena and contains the last four lessons: 27 architectural acoustics, i.e. it investigates the degree to - Perception of space, rooms, and distance; 28 - which the acoustic simulations of a closed space agree Introduction to the soundscape approach; 29 - Perceptual with actual measurements. The second practical case evaluation of noise; 30 - Auditory illusions. exhibits how the binaural transfer path synthesis The two practical cases give examples of how the technique can be applied to a car simulator. presented concepts are used practical, real-life applications. The first case presents the process of 3.4. Electroacoustics producing aurally correct recordings to be used in listening tests designed for sound quality evaluation. The Electroacoustics is a specific field as it involves both second case presents the design of a listening experiment electrical engineering and acoustics. Moreover, it ties that investigates the localization of sound sources. together the electrical, mechanical, and acoustical domain. 3.3. Acoustic simulations and auralization The course is divided into seven sections, and the first one is a refresher of sorts, as it covers the background This course covers the field of acoustical simulations and knowledge required to successfully follow this course via auralization that stems from them and has proven to be these lessons: 1 - Introduction to electroacoustics; 2 - an invaluable tool that helps us gain understanding about Basics of electricity; 3 - Mechanics: damped harmonic a range of phenomena by direct perception (listening). oscillator; 4 - Acoustical circuits; 5 - Mechanical- The first section offers an introduction to this field via five acoustical-electrical analogies; 6 - Radiation of simple lessons: 1 - Introduction and overview of acoustical acoustic sources. simulation; 2 - Impulse responses; 3 - Convolution; 4 - The second section is focused on the characterization of Fourier transform; 5 - Discrete processing. audio systems through two lessons: 7 - Linear The second section focuses on the fundamentals of room characteristics of transducers; 8 - Transducer limitations. acoustic simulations and consists of the following lessons: The third section introduces commonly used transduction 6 - Geometrical acoustics; 7 - Prediction of reverberation principles and consists of three lessons: 9 - Electrodynamic time; 8 - Image source model; 9 - Ray tracing; 10 - transduction; 10 - Electrostatic transduction; 11 - Radiosity methods; 11 - Wave-based models. Acoustical-mechanical transduction. The third section introduces the models used in noise The fourth section deals with the modelling of transducers control and sound design through the following lessons: via the following lessons: 12 - Electromechanical source: 12 - Environmental sound propagation; 13 - Airborne the shaker; 13 - Electromechanical sensor: the geophone; sound in buildings; 14 - Impact sound in buildings; 15 - 14 - Directive microphones; 15 - Electrodynamic Binaural transfer path analysis. microphone; 16 - Electrostatic microphone; 17 - The fourth section is oriented on auralization and Unidirectional electrodynamic microphone; 18 - introduces this technique via these lessons: 16 - Binaural Electrodynamic moving coil loudspeaker; 19 - Thiele & synthesis; 17 - Synthesis of room impulse responses; 18 - Small parameters: theory and measurement. FIR convolution; 19 - Real-time convolution. The fifth section illustrates the most common loudspeaker The fifth section presents the techniques of sound enclosures and systems and consists of the following reproduction and consists of these five lessons: 20 - lessons: 20 - Closed box system; 21 - Vented box system; Headphones; 21 - Binaural loudspeaker reproduction; 22 - 22 - Other types of enclosures; 23 - Multichannel speaker Stereo and vector-base amplitude panning; 23 - Panning systems. techniques and object-based audio; 24 - Ambisonics. The sixth section explains advanced modelling techniques The last section discusses the applications and evaluation of loudspeaker systems and is composed of three lessons: of simulations via these lessons: 25 - Architectural design; 24 - T&S model limits; 25 - Advanced parameters 26 - Noise mapping; 27 - Application of simulation and estimation; 26 - Loudspeaker nonlinearities. Horvat et al.: Acoustics Knowledge Alliance project: the development of open-access interactive online educational materials in acoustics - strategy and results 169 AAAA – 2023 – IZOLA - Conference Proceedings The final section deals with multi-transducer designs via 23 - Acoustic classification of sound insulation; 24 - these lessons: 27 - Microphone arrays; 28 - Loudspeaker Examples of good and bad constructions and arrays; 29 - Personal sound zones; 30 - Playback systems. workmanship practice; 25 - Influence of absorption on The first practical case demonstrates the design of a two- sound insulation (or how room acoustics influences way loudspeaker system. The second practical case shows building acoustics); 26 - Materials in practice; 27 - Towards the measurements that are made on a loudspeaker driver. more sustainable and low-energy produced materials for buildings; 28 - Special construction types using vibration 3.5. Room and building acoustics isolation; 29 - Background noise and the sources contributing to it; 30 - Practical approaches for innovative, This course deals with the overall acoustic comfort of sustainable, and cost-effective buildings. closed spaces by addressing its three key components. As acoustic comfort depends on all three of its The first one is room acoustics, which strives to provide components, the two practical cases are joined into one acoustic conditions within a room that suit its size and that deals with acoustic comfort in a comprehensive purpose. The second one is building acoustics, which manner, as illustrated with a number of practical provides protection from sound coming from adjacent examples embedded into the structure of the practical spaces, but also protects these spaces from sound case. The intent is to display how the principles of room generated in the room. The third one is internal noise and building acoustics are used in real-life situations on emitted by noise sources within the room itself. different building designs. The course is divided into two large sections. The first one covers room acoustics, and the second one deals with building acoustics and noise. Since there is a commonly 4. SELECTED EXAMPLES OF DEVELOPED MATERIALS spread confusion and misconception about the difference between room acoustics and building acoustics, the This section displays some examples of the developed course starts with an introductory lesson: 1 - What is the educational material. The material is presented in the difference between room acoustics and building ready-to-be-published form, after being redesigned by acoustics? the graphic designers. As the interactive components The section on room acoustics contains the following cannot be displayed, the material is presented as images lessons: 2 - How a room responds to sound; 3 - Acoustic that depict only the layout and static content. The full treatment - absorption, reflection, and diffusion; 4 - interactivity of the courseware material can be explored Volume and shape of the room and their influence on on the ACOUCOU platform site [1,5]. The theory parts room acoustics; 5 - Reverberation time; 6 - How a sound (part A) of the lessons are not shown here, as they are of source interacts with a room; 7 - How loudspeakers classic design, i.e. based on text complemented with interact with rooms; 8 - Acoustic elements - absorbers; 9 - figures, tables and equations. Instead, examples of the Acoustic elements - diffusers; 10 - General requirements principle (part B) and task (part C) parts have been chosen for good acoustics; 11 - Objective and subjective from lessons in all five courses, as they reflect the parameters in room acoustics; 12 - Room and building creativity displayed by their authors during the design acoustics design criteria; 13 - Single-speaker and multi-process. speaker environments; 14 - Acoustics of spaces for music; Examples of the principle parts of selected lessons are 15 - Room acoustic design elements at the disposal of shown in Figures 4 to 8, while the task parts of the architectural design. selected lessons are displayed in Figures 9 to 13. The section on building acoustics (and noise) is composed The principle part shown in Figure 4 allows the user to of these lessons: 16 - Airborne and impact sound, direct choose between a shoebox, cylindrical, and spherical and flanking transmission; 17 - Sound reduction index for cavity of predefined dimensions. The pressure distribution simple and composite structures; 18 - Single-leaf and of different modes can be observed both on the double-leaf walls; 19 - Floors and ceilings; 20 - Facades: boundaries of the cavity and inside it by changing the curtain walls; 21 - Doors and windows; 22 - Measurement number of nodes in appropriate axes. The frequency of and evaluation of airborne and impact sound insulation; the mode is displayed as well. Horvat et al.: Acoustics Knowledge Alliance project: the development of open-access interactive online educational materials in acoustics - strategy and results 170 AAAA – 2023 – IZOLA - Conference Proceedings Fig.4. The principle part of the lesson on modes in a Fig.6. The principle part of the lesson on convolution cavity The principle part shown in Figure 7 illustrates the The principle part shown in Figure 5 allows the user to modelling of the acoustic circuit that represents an open, perceive the width of Bark bands (critical bands) by empty wine bottle. The bottle is first modelled using a listening. At low frequencies, a shift of 100 Hz is equal to cavity (body of the bottle) with an elongated opening 1 Bark, whereas the same shift at high frequencies (from (bottleneck), and exact dimensions of the two parts are 6400 to 6500 Hz) is equivalent to only 1/13 of a Bark. measured and input into the model. Using the appropriate The principle part shown in Figure 6 illustrates the formulae, one can calculate the theoretical resonance technique of convolution. The user can change the frequency of the bottle. The actual resonance frequency reverberation time of an imaginary room, and a of the bottle is measured and recorded, and this sound corresponding impulse response is generated through can be listened to. At the same time, the spectrum of that simulation. The impulse response of the room is sound can be checked, and it clearly shows the actual convolved with a dry audio signal. The resulting audio resonance frequency. Finally, length corrections are signal sounds as if it was recorded in the simulated room, applied to the measured length of the bottleneck to and the user can listen to both the original sound and the account for the abrupt changes of the cross-section of the convolved sound and compare them, thus evaluating the bottleneck at its ends, resulting in a considerably effect of convolution. improved prediction of the resonance frequency. Fig.5. The principle part of the lesson on bark bands Fig.7. The principle part of the lesson on acoustical circuits Horvat et al.: Acoustics Knowledge Alliance project: the development of open-access interactive online educational materials in acoustics - strategy and results 171 AAAA – 2023 – IZOLA - Conference Proceedings The principle part shown in Figure 8 allows the user to study the construction of porous and resonant absorbers by changing their configuration. The changes are illustrated by an image of the actual construction, and the calculated frequency-dependent sound absorption coefficient. In case of resonant absorbers shown here, the user can change the thickness and the percentage of perforation of the perforated panel, the thickness of the porous panel, the total width of the gap between the perforated panel and the base wall, and the flow resistivity of the porous material. Each of the five parameters has two distinct values, which enables the user to examine 32 different cases. Fig.9. The task part of the lesson on resonance in a cavity Fig.8. The principle part of the lesson on absorbers The task part shown in Figure 9 requires the user to figure Fig.10. The task part of the lesson on loudness out which modes in the illustrated cavity (two- dimensional for simplicity) will be excited by the source placed at the marked position (orange dot), and which will not be excited. Five different source placements are given for the same cavity. The task part shown in Figure 10 requires the user to answer to a question regarding the dependence of loudness on different parameters and effects. This type of tasks is based on multiple-choice questions with one or more correct answers. In this case there are five correct answers. The task part shown in Figure 11 requires the user to examine a simple mass-spring system that consists of a mass, a spring, and a damper, and is excited with Dirac excitation at t = 0. The user needs to determine the correct Fig.11. The task part of the lesson on (simulations) shape of the impulse response of that system. Horvat et al.: Acoustics Knowledge Alliance project: the development of open-access interactive online educational materials in acoustics - strategy and results 172 AAAA – 2023 – IZOLA - Conference Proceedings The task part shown in Figure 12 challenges the user to 5. CONCLUSIONS determine the loudspeaker construction, among the four that are offered, which exhibits a bidirectional radiation At this moment, the Acoustics Knowledge Alliance pattern. The parameters to keep in mind are the physical (ASKNOW) project, funded by the Erasmus+ Knowledge layout of the loudspeakers, the polarity of the feed signals Alliances programme of the EU, has officially ended. The (for constructions with two loudspeakers), and the results of the project reflect the tremendous effort put in relative frequency range of operation determined with by the project consortium over the past three and a half the relationship between the physical dimensions of the years. The consortium was composed as a symbiosis of the loudspeakers and the wavelength of the emitted sound academic and the business world and aided by outside wave. partners. Within the scope of the project, the consortium In the task part shown in Figure 13, the user must identify has successfully developed freely available online architectural elements designed to modify the acoustics educational materials under the CC BY-NC-ND 4.0 license of a room and sort them into appropriate categories (Creative Commons license) that cover five different according to their dominant behaviour as absorbers, topics: acoustic fundamentals, psychoacoustics, acoustic reflectors, diffusers, or sound-transparent elements. simulations and auralization, and room and building acoustics in as many correspondingly named courses. The developed educational materials are infused with diverse forms of interactive content, thus enhancing the learning experience for the users, and making the process of knowledge transfer both engaging and effective. The primary target groups for which the materials have been developed are students and educators, as well as professionals in the field. As such, the content of the materials is not necessarily trivial, but it is the strong belief of the entire project consortium that they can be used by anyone who already has a basic understanding of acoustics and is interested in improving their knowledge. Fig.12. The task part of the lesson on the radiation of 6. ACKNOWLEDGMENTS simple acoustic sources All the activities within the Acoustics Knowledge Alliance (ASKNOW) project (project reference: 612425-EPP-1- 2019-1-FR-EPPKA2-KA) have been funded by the Education, Audiovisual and Culture Executive Agency (EACEA) through the ERASMUS+: Knowledge Alliances programme. The authors would like to thank all the external reviewers who took their time to review the developed materials and have provided invaluable feedback to the project consortium required for improving the quality of the final product. The authors would also like to thank the volunteers who took part in testing the developed materials for functionality and accuracy. Their feedback has proved to Fig.13. The task part of the lesson on acoustics of spaces be crucial for final improvements of the developed for speech educational materials. Horvat et al.: Acoustics Knowledge Alliance project: the development of open-access interactive online educational materials in acoustics - strategy and results 173 AAAA – 2023 – IZOLA - Conference Proceedings 7. REFERENCES [7.] Horvat, M., Jaruszewska, K., Raetz, S., Carayol, E., Sluyts, Y., Herweg, A., Aspöck, L. and Zeman, S. The [1.] The homepage of the ACOUCOU platform. Available at: development of modern, interactive acoustic https://acoucou.org (Accessed on 10 July 2023) courseware material within the Acoustics Knowledge Al iance project, in Proceedings of the Euroregio – [2.] Jaruszewska, K., Barański, F., Piotrowska, M., Melon, M., BNAM 2022 Joint Acoustics Conference, pp. 225-234, Dazel, O., Vorländer, M., Aspöck, L., Horvat, M., 2022. Jambrošić, K., Rychtáriková, M., Kritly, L. and Herweg, A. ACOUCOU Platform to Acquire Professional Skil s and [8.] Melon, M., Sluyts, Y., Heimes, A., Horvat, M., Knowledge in the Field of Acoustics., in Proceedings of Jaruszewska, K., Herweg, A., Wojtyła, B. and Dalmont, J.- the 23rd International Congress on Acoustics, pp. 4348- P. Open access online course in acoustics in the frame 4355, 2019. of AcouStics Knowledge al iance (ASKnow) project” in Proceedings of the 16ème Congrès Français [3.] Jaruszewska, K., Melon, M., Dazel, O., Vorländer, M., d’Acoustique, pp. 1-6, 2022. Aspöck, L., Wulfrank, T., Jambrošić, K., Horvat, M., Rychtáriková, M., Sluyts, Y., Wojtyla, B., Herweg, A. and [9.] Jambrošić, K., Aspöck, L., Carayol, E., Herweg, A., Horvat, Barański, F. ACOUCOU - online platform including M., Jaruszewska, K., Novak, A., Sluyts, Y. and Wojtyla, B. interactive educational materials about acoustics, in The Acoustics Knowledge Al iance project: the most Proceedings of Forum Acusticum 2020, pp. 3023-3024, recent addition to the Acoustic Courseware online 2020. educational platform, in Proceedings of Inter-Noise 2023, 2023. [4.] Jaruszewska, K., Melon, M., Dazel, O., Vorländer, M., Rychtáriková, M., Horvat, M., Wulfrank, T., Herweg, A., [10.] Jaruszewska, K., Novak, A., Dazel. O., Melon, M., Sluyts, Aspöck, L., Sluyts, Y., Jambrošić, K., Carayol, E., Wojtyla, Y., Heimes, A., Jambrosic, K., Kruh-Elendt, A., Carayol, E. B., Łuczak, M. and Chmelík, V. The ACOUCOU platform: and Zeman, S. Approach to developing online Online acoustic education developed by an educational materials based on Acoustics Knowledge interdisciplinary team, The Journal of the Acoustical Al iance (ASKNOW) project, in Proceedings of Forum Society of America, 152(3), pp. 1922-1931, 2022. Acusticum 2023, 2023. [5.] The homepage of the ASKNOW project. Available at: [11.] Horvat, M., Raetz, S., Lotton, P., Sluyts, Y., Vorländer, https://asknow.acoucou.org (Accessed on 10 July 2023) M., Petošić, A., Jaruszewska, K., Ray, A., Wojtyla, B. and Wuflrank, T. Open-access interactive online [6.] Horvat, M., Melon, M., Jaruszewska, K., Wulfrank, T., courseware in acoustics developed within the scope of Sluyts, Y., Herweg, A., Aspöck, L. and Wojtyla, B. the acoustics knowledge al iance (ASKNOW) project, in Acoustics Knowledge Al iance (ASKnow) project as the Proceedings of Forum Acusticum 2023, 2023. latest addition to the Acoustic Courseware (ACOUCOU) learning platform, in Proceedings of the 9th Congress of the Alps-Adria Acoustics Association, pp. 160-168, 2021. Horvat et al.: Acoustics Knowledge Alliance project: the development of open-access interactive online educational materials in acoustics - strategy and results 174 Keynote Invited speech Challenges in sound insulation of wooden buildings Prof. Dr. Jonas Brunskog Department of Electrical and Photonics Engineering, Technical University of Denmark (DTU) E-mail: jbru@dtu.dk Dr. Jonas Brunskog is Associate Professor at the Acoustic Technology group at the Department of Electrical and Photonics Engineering, at Technical University of Denmark (DTU). He is a co-author of 55 scientific journal articles, and more than 110 conference publications. His works have been cited 1960 times (Google Scholar). He has supervised 10 completed PhD theses as main supervisor, and 8 as co-supervisor. Scientific focus areas are: General acoustics, Vibro-acoustics, Building acoustics, Room acoustics, Numerical acoustics, Environmental acoustics, Signal processing, and Voice research. Adequate sound insulation is important in buildings due to legal, comfort and health reasons, the latter emphasized by WHO. The sound insulation of a single homogenous wall is mainly given by the mass per unit area, the mass law, leaving not much room for improve-ment. Economic pressure as well as environmental demands of increased use of wood in buildings leads to reduced weight, which thus might resulting in poor sound insulation. However, there are also certain aspects that talk in favor for the use of wood constructions. In his lecture, Dr. Brunskog will review the acoustic challenges and possibilities when building with timber constructions. Timber constructions are used in both old traditional floor constructions in many countries, and in new innovative buildings with high sound insulation. Most traditional as well new timber floor constructions are spatially periodic, which will lead to periodic effects such as pass- and stop bands. In other areas of physics and engineering, such aspects are used as metamaterials. Another branch of constructions where innovative ideas are needed and are being used are CLT constructions. 175 Technical keynote Invited speech Design and construction of the InnoRenew CoE Acoustics Laboratory Assist Prof. Dr. Rok Prislan InnoRenew CoE E-mail: rok.prislan@innorenew.eu Dr. Rok Prislan is research group leader for Buildings and is leading research in the field of acoustics at InnoRenew CoE. Dr. Prislan is also holding the position of an assistant professor at the University of Primorska, where he teaches physics. His main research topics are advanced measurement techniques for sound field characterization and geometrical room acoustic modelling. As part of his career, dr. Prislan was working as an acoustic designer/consultant and has led over 60 projects in the field of acoustics and noise control. Dr. Prislan has two master’s degrees, one in physics (mathematical physics) from the University of Ljubljana and one in engineering acoustics from the Technical University of Denmark (DTU). InnoRenew CoE was established in 2017 as an international research institute with the goal of providing innovation and conducting cutting-edge research in the broad field of sustainable and healthy built environment. In just five years, the institute was built from the ground up by assembling an international team of researchers, investing in state-of-the-art laboratory equipment, and constructing the laboratories. The acoustics laboratory was the most technically challenging laboratory to design and build. In this talk, dr. Rok Prislan, who played a leading role in the design, will present this process. For the newly built reverberation and anechoic chambers, dr. Prislan will provide an overview of the given requirements, explain how the chambers were conceived, what the main design challenges were, and address the uncertainties that arise from the undefined guidelines for the design of such facilities. Dr. Prislan also explains the approaches used in this engineering adventure, including the acoustic modeling methods used to predict and the experimental methods used to evaluate the performance of the chambers. The presentation of the main construction steps is supported by rich photographic and video material. 176 Contributed papers Building acoustics 1. Design and construction of a temporary test facility for sound insulation measurements of doors Nika Šubic (MK3 d.o.o.) 2. Acoustic performance of buildings, components and materials as a parameter for ecological and social sustainability assessments Franz Dolezal (IBO - Austrian Institute for Building and Ecology) 3. Improvement of impact sound insulation with tile underlay materials – impact sound insulation without floating floors Beáta Mesterházy (Department of Acoustics of OPAKFI, BME (Budapest University of Tecnology and Economics), Department of Building Constructions, Laboratory of Building Acoustics) 177 DESING AND CONSTRUCTION OF A TEMPORARY TEST FACILITY FOR SOUND INSULATION MEASUREMENTS OF DOORS Nika Šubic(1), Andrej Šubic(1), Zdenka Šubic(1), Martin Klun(1), Rok Prislan(2) (1)MK3 d.o.o., (2)InnoRenew CoE Abstract: To fulfill the sound insulation requirements for separating elements in buildings a common concern is reaching sufficient sound insulation of smaller construction elements, such as doors. Besides the construction of the door leaf, achieving quality sealing of the door leaf with the frame and floor is important as well as the on-site execution, which can greatly impact the final sound insulation of the whole separating element. To optimize the design of doors regarding sound insulation, repetitive testing is required which can be time consuming and expensive when performed on a different location by an external contractor. To facilitate the process for a Slovenian company, a temporary test facility was designed on their premises. The test facility was constructed in the existing building of the company using lightweight materials. The design and construction followed the requirements of the ISO 10140 standard, and the finished test facility can test door specimens with sound reduction index values of up to 56 dB. Keywords: sound insulation, test facility, measurement, doors 1. INTRODUCTION The paper presents the design and the acoustic characteristics of the completed test facility. Achieving sufficient sound insulation of doors in buildings is a in a common challenge, the main complication usually being inadequate sealing of the door leaf with the frame 2. DESIGN OF THE TEST FACILITY and the floor, as well as proper installation of the door frame in the separating wall. To improve the existing The design of the temporary test facility follows the design of the doors and gain experience with the guidelines of the ISO 10140-5 standard [1], with the installation procedure a design of a temporary test facility distinction of a lightweight construction. The basic for sound reduction index measurements was proposed. principle of the construction was to use a room-in-room It was required that the new construction was lightweight design for the receiver room, while using the existing since it was placed on the 1st floor of an existing building premises as the source room, for which minimal with structural limitations to sustain excessive loads such interventions are needed. as of a concrete structure. Additionally, the lightweight construction can be more easily disassembled when it is 2.1. Existing structure, placement, and geometry of the not needed anymore. It is difficult to achieve good low- test facility frequency sound insulation with lightweight construction however, it is equally uncommon to expect high values of The building in which the test facility is placed is mainly of sound reduction index at low-frequency for small massive construction. The walls are 18-20 cm thick and construction elements. For this reason, it was foreseen made of concrete, the floor is a prefabricated concrete that a lightweight test facility could be an appropriate slab with approximate thickness of 25 cm. The surface solution for measurements of smaller construction mass of all the separating walls and floor is at least 400 elements such as doors and windows. kg/m2. The existing roof construction is a lightweight Šubic et al.: Design and construction of a temporary test facility for sound insulation measurements of doors 178 AAAA – 2023 – IZOLA - Conference Proceedings metal sandwich construction with a suspended ceiling shell was designed as an additional lightweight with a surface mass of approximately 45 kg/m2. However construction. Wals A, B and the ceiling were designed as the roof is not a part of the test facility structure as a a double shell lightweight partition. secondary ceiling was designed and constructed. The floating floor design had an 8 cm wooden structure The new test facility consists of two adjacent rooms with OSB board planking and sand as a filling material. The situated at a far end of the existing premises, as shown in surface mass of the floating floor was at least 135 kg/m2. Fig.1 and Fig.2. The source room has a net volume of 75,5 The elastic layer was implemented in strips on the boarder m3 and the receiver room a net volume of 65,5 m3 – both of the floating floor – under the inside shell of the walls, rooms fulfil the ISO 10140-5 standard [1] requirement of and under the central part. The thickness of the material a minimal volume of 50 m3 and a difference in volumes of and size of the strips was selected according to the surface at least 10%. pressure of the floating floor, walls and ceiling weight in the specific area, with the goal of achieving a resonant frequency under 15 Hz. The finished floating floor hight was about 20 cm. Walls A and B were designed as a double shell lightweight partition, with the inside shell placed on the floating floor and the outside shell placed on the existing concrete slab. The outside shell was designed with two layers of drywall and one layer of OSB planking with the surface mass of 29 kg/m2 on a metal substructure. The inside shell had three layers of drywall and one layer of OSB planking with the Fig.1. Floorplan of the test facility: The existing structure surface mass of 38 kg/m2 on a metal substructure. The 20 is shown in grey and the new walls in blue. cm cavity was filled with mineral wool of density 45 kg/m3. The inside shell of walls C and D was designed in the same way as the inside shell of walls A and B, but with a smaller cavity thickness of 7,5 cm (metal structure filled with mineral wool) + 3 cm air gap towards the existing outer walls. The ceiling construction also followed a double shell design, where the outside shell was supported by the outside shell of the walls and the inside shell was completely supported by the inside shell of the walls. Again, the outside shell had three layers of planking (two layers of drywall and one layer of OSB plates) and the Fig.2. Cross section of the test facility: The existing inside shell four layers of planking (three layers of drywall structure is shown in grey and the new structure in blue. and one layer of OSB plates). The cavity was 17,5 cm thick and filled with mineral wool of density 45 kg/m3. 2.2. New partition elements The source room was mainly closed off by the existing construction. An additional barrier was designed towards The design of the test facility was mainly focused on the rest of the hall to limit the volume of the room. achieving sufficient sound insulation towards the Considering the mass law equation (ISO 12354-1 receiving room. That is why a room-in-room design was standard) [2] the minimal weighted sound reduction index proposed taking full advantage of the existing massive ( Rw) of the existing structure is Rw = 56 dB. With the construction. A separated double shell construction is proposed additional layers, the Rw of walls C and D is 70 designed for all separating elements of the receiving dB. Regarding walls A, B and the ceiling partitions, a similar room. sound insulation rating of Rw = 68 dB is expected [3]. For walls C, D (Fig.1.) and the floor (Fig.2.) the existing construction served as the outside shell, while the inside Šubic et al.: Design and construction of a temporary test facility for sound insulation measurements of doors 179 AAAA – 2023 – IZOLA - Conference Proceedings 2.3. Opening for door specimens The door opening was designed in a way that it can be adjusted in size; however, it is most commonly used for a door leaf of the dimension of 1 m x 2,1 m. Additional 5-10 cm are usually needed on each side of the opening for the door frame installation. The conjunction between the wall and the test element was designed in a way, to prevent flanking transmission and sound leakage in the greatest way possible. Two layers of OSB boards and one layer of drywall sheets was used, screwed together over a flexible sound insulation material. If the size of the opening is adjusted for a specific door element, the detail of the conjunction has to be replicated. 3. RESULTS AND DISSCUSION Fig.3. Measured values of reverberation time in the All measurements were executed in accordance with ISO empty receiver room, which are inside the required 10140-4 standard [4] in the frequency range 100 Hz – 5 range at all relevant frequencies. kHz. The measurement methods used fixed microphone positions. Background noise level was more than 15 dB 3.2. Sound insulation of the test facility below the generated sound at all relevant frequencies. One person was present in the room during the To assess the maximal sound insulation of the test facility measurements. an additional multi-layer covering was applied over the door opening. The covering was of a similar composition 3.1. Reverberation time of the receiver room as the wall to which it was tightly sealed. The airborne sound insulation was measured in accordance with ISO The reverberation time was measured in accordance with 10140-4 [4] and ISO 10140-5 [1] standards. The single ISO 10140-4 [4] and ISO 3382-2 [5] standards. Engineering value weighted sound reduction index ( Rw) and the method was applied with two source positions and four spectrum adaptation terms C and Ctr were calculated in microphone positions. According to ISO 10140-5 standard accordance with ISO 717-1 standard [6] considering the [1] the reverberation time over the evaluated frequency whole wall as the separating structure. Fig.4. presents the range is limited with equation (1), measured values of sound reduction index of the test facility over the relevant frequency range, with the 1 ≤ T ≤ 2(V/50)2/3 [s] (1) corresponding single number value at Rw ( C, Ctr) = 62 (-2, - 7) dB. where T represents the reverberation time and V the As expected for lightweight structure, the reduction index volume of the receiver room. In the specific case the is below the reference curve at lower frequencies and reverberation time is in the range of 1 s ≤ T ≤ 2,37 s. It is above it at higher frequencies. No insulation drops usually evident form the measurement results shown in Fig.3. , associated with structural resonances can be observed. that the reverberation time of the receiver room is inside In accordance with the limitations regarding flanking the required range at all relevant frequencies. transmission (Annex A, ISO 10140-2 [7]) the sound reduction index of the measured element (door) must be at least 6 dB lower than the sound reduction index of the test facility at all frequencies. Therefore, the test facility permits testing of doors with Rw value up to 56 dB. Šubic et al.: Design and construction of a temporary test facility for sound insulation measurements of doors 180 AAAA – 2023 – IZOLA - Conference Proceedings commonly face challenges which have been systematically investigated since the 70's by Kihlman and Nilsson [8]. It has been also shown by Craik [9] that inter- laboratory deviation can originate from several design differences between facilities, which are not even specified by the standardized measurement procedures. Fig.4. Measured values of maximal sound reduction index R of the test facility with the relevant reference curve (717-1). 3.3. Sound insulation of tested doors More than 70 tests of doors were conducted until now in order to test different compositions, door elements and Fig.5. Measured values of sound reduction index of a sealing options. The maximal measured sound insulation specimen ( R1) in comparison with the official results ( R2) of the tested specimens reached R and the relevant reference curve for R1 values. w ( C, Ctr) = 50 (-1, -4) dB. One of the specimens was also sent for official testing to a certified laboratory with intention of verifying the test R1 R2 result. The measured frequency dependent values (Fig.5.) Rw ( C, Ctr) 49 (-2, 5) dB 50 (-2, -6) dB are very comparable, with the official results ( R2) showing a slightly higher sound reduction at high frequencies. Table.1. Comparison of the measured ( R1) and official Consequently, the Rw value is slightly higher for the R2 ( R2) Rw values. measurement (Table.1.). Excluding the frequency range above 1000 Hz, the standard deviation between measurements in each frequency band is less than 1,5 dB, 4. CONCLUSION while the maximal difference is 5 dB. The lower sound reduction values at high frequencies, A temporary test facility for sound insulation measured at our test facility could be a result of leakage measurements was designed and built to support the at the installation of the door frame which is a common development of high sound insulating doors. It enables door installation failure. In addition, interlaboratory testing of doors with sound insulation up to Rw = 56 dB, variations are expected to a certain degree, which is also with Rw ( C, Ctr) = 50 (-1, -4) dB being the highest measured the reason that standardized measurements procedures value until now. The validity of test results was confirmed are constantly evolving and improving. Testing facilities Šubic et al.: Design and construction of a temporary test facility for sound insulation measurements of doors 181 AAAA – 2023 – IZOLA - Conference Proceedings with results of a certified laboratory, with the small single elements – Part 4: Measurement procedures and number rating deviation of 1 dB. Additional tests requirements, International Organization for Standardization, 2010. comparing the facility to certified laboratories would be [5.] ISO 3382-2:2008. Acoustics — Measurement of room beneficial to further investigate the reason behind the acoustic parameters — Part 2: Reverberation time in deviations observed at higher frequencies. ordinary rooms, International Organization for Standardization, 2008. 5. REFERENCES [6.] ISO 717-1:2020. Acoustics — Rating of sound insulation in buildings and of building elements — [1.] ISO 10140-5:2010. Acoustics – Laboratory Part 1: Airborne sound insulation, International measurement of sound insulation of building Organization for Standardization, 2020. elements – Part 5: Requirements for test facilities [7.] ISO 10140-2:2010. Acoustics – Laboratory and equipment, International Organization for measurement of sound insulation of building Standardization, 2010. elements – Part 2: Measurement of airborne sound [2.] ISO 12354-1:2017. Building acoustics – Estimation of insulation, International Organization for acoustic performance of buildings from the Standardization, 2010. performance of elements – Part 1: Airborne sound [8.] T. Kihlman, A.C. Nilsson, The effects of some insulation between rooms, International laboratory designs and mounting conditions on Organization for Standardization, 2017. reduction index measurements, Journal of Sound [3.] SIA, Bauteildokumentation Schallschutz im Hochbau and Vibration, Volume 24, Issue 3, pp. 349-364, 1972. - Zusammenstellung gemessener Bauteile, SIA [9.] Robert J.M. Craik, The influence of the laboratory on Zurich, 2005. measurements of wall performance, Applied [4.] ISO 10140-4:2010. Acoustics – Laboratory Acoustics, Volume 35, Issue 1, pp. 25-46, 1992. measurement of sound insulation of building Šubic et al.: Design and construction of a temporary test facility for sound insulation measurements of doors 182 ACOUSTIC PERFORMANCE OF BUILDINGS, COMPONENTS AND MATERIALS AS A PARAMETER FOR ECOLOGICAL AND SOCIAL SUSTAINABILITY ASSESSMENTS Franz Dolezal1, Maria Fellner1, Maximilian Neusser2 1IBO – Austrian Institute for Building and Ecology, 1090 Vienna, Austria 2Research Division for Building Physics, Technische Universität Wien, 1040 Vienna, Austria Abstract: Several initiatives have been launched to increase sustainability in the building industry. The consequence was a broader view of sustainability, not only based on the environmental, but on the economic and social performance of materials, components and buildings as well. A closer look on internationally acting Green Building Labels reveals that acoustic performance is, although a technical quality, handled as part of the social sustainability aspects of a building. Depending on the label, several credits are assigned for the fulfilment of requirements. Particularly for sustainable lightweight structures made of wood, sound insulation is to some extent defined by the properties of the insulation material, filled into the cavity of the timber joists. Especially the density and the dynamic stiffness, as well as the air flow resistivity, are important material parameters to achieve satisfying sound insulation properties. A comparison with different products and components has been carried out. It could be shown that the use of renewables leads to lower environmental impacts, though, the insulation material is not the most relevant parameter of the building components. Keywords: Sustainability, acoustics, sustainable building assessment, environmental performance, acoustic performance 1. INTRODUCTION assessment, developed by CEN/TC 350 “Sustainability of construction works” since 2005 according the European The building sector plays a major role when it comes Commission’s mandate [3] with all aspects of to greenhouse gas (GHG) Emissions and tackling of climate sustainability included. Based on this strategy, building change. According to [1], almost 40% of energy-related assessment and rating schemes offer solutions to greenhouse gas emissions come from the building and achieving transformation to a more sustainable building construction sector. The importance of a more sector, including the economic and social column of pronounced focus on sustainability in the building sustainability. Interestingly building physical aspects like industry is proved by the fact that buildings and structures acoustic properties of components and noise protection, need 50 % of all natural resources utilized and create according to EN 16309 [4], is assigned to the social aspects around 60 % of all wastes, as a consequence of their as shown in figure 1. production, building, usage and maintenance [2). Current Considering climate crises and resulting UN Sustainable holistic approaches in the construction sector include Development Goal number 13 [5] as well as the European economic and social aspects as well. According to the Green Deal [6], environmental impact of building general definition of sustainability, ecological, economic components becomes more and more important. and social aspects and consequences – also called the Moreover, since the draft of the European Construction “three columns of sustainability” – are taken into account Product Regulation [7] indicates the assessment of the to analyse and assess buildings and civil engineering environmental impact as a basic requirement (sustainable works. This is already pictured in relevant voluntary, use of natural resources of construction work) with a list horizontal standards for the sustainable building of essential characteristics related to life cycle assessment Dolezal et al.: Acoustic Performance as a Parameter for Ecological and Social Sustainability Assessments 183 AAAA – 2023 – IZOLA - Conference Proceedings (LCA) to be covered, development is gaining traction and as an important issue of sustainability, which acoustic awareness of the industry regarding sustainability is rising. characteristics are considered and which methodological approaches are used for rating. Acoustical aspect Calculation Measurement Airborne sound EN 12354-1 EN ISO 16283-1 Impact sound EN 12354-2 EN ISO 16283-2 Sound from outside EN 12354-3 EN ISO 16283-3 Service equipment EN 12354-5 Room acoustics EN 12354-6 EN ISO 3382-2 Table 1. Recommended acoustical aspects and normative assessment methods acc. to EN 16309 Analysis is based on the web pages and publications on criteria, assessment and credits. To find out the impact of Fig.1. Three columns of sustainability with acoustic acoustic properties on SBAS`s rating results, the different indicators shown as part of social sustainability aspects in assessment schemes had to be investigated. EN 16309 (right column) Only the most important acoustic aspects concerning airborne and impact sound insulation have been Though noise protection in buildings is seen as part of the considered. Moreover, focus was laid on schemes related social sustainability, sound insulation of components is to residential buildings have been taken into account. mainly defined by their structure and the materials used. Selection of materials can have a significant impact on 2.1. Level(s) environmental sustainability and sound insulation properties as shown in [8] and with focus on the regional, Level(s) is a common framework for sustainable buildings renewable alternative straw as building material in [9]. which was launched by the European Commission in 2015 to improve the performance of buildings towards reduced carbon emissions, higher material and resource efficiency, 2. SUSTAINABLE BUILDING ASSESSMENT SCHEMES health and well-being as wells as adaptations to climate change impacts, taking into account the whole life cycle of Social sustainability assessment of buildings is related to a buildings. Level(s) offers open source manual and tools variety of different issues. Concerning acoustics, (beta-versions were relaunched and tested between 2015 according to EN 16309, acoustic characteristics are and 2020). The assessment system corresponds with the defined by airborne and impact sound insulation of goals of the EU Green Deal [6], new Circular Economy separating walls and floors, sound insulation of the Action Plan [10] and the Paris Agreement to significantly external envelope, noise level including service reduce the carbon emissions in the construction industry equipment noise, and reverberation time. Furthermore it by 2050 [11]. is mentioned that different types of use shall be taken into Acoustics and noise protection are defined as one of 16 account. A detailed overview of recommended aspects core indicators, which have to be evaluated within three and standards for evaluation is given in table 1. Since EN levels: level 1 the conceptual design phase, level 2 the 16309 serves as a methodological basis for Sustainable detailed design stag, level 3 requires measurements to Building Assessment Schemes (SBAS) most of them try to assess the as built performance classified by five consider all of these aspects. subcategories: façade, airborne and impact sound Investigated were schemes from German speaking insulation, service equipment and sound absorption countries (DGNB, ÖGNI, IBO-Ökopass, BNB) as well as the where applicable. Core Indicators, categories or sub- ones with international importance (Level(s), BREAM). categories have no weighting to each other within the The objective was to figure out, if sound protection is seen Level(s) building assessment system. With regard to target Dolezal et al.: Acoustic Performance as a Parameter for Ecological and Social Sustainability Assessments 184 AAAA – 2023 – IZOLA - Conference Proceedings values, Level(s) does not define its own requirements, but Additionally, minimum requirements for project support refers to national building regulations or classification have to be fulfilled: an appropriately qualified sound systems. Rating procedures are defined in accordance to insulation / acoustic planner is appointed by the client at ISO 717-1 [12] for airborne sound insulation and ISO 717- a suitable stage of the planning process to determine 2 [13] for impact sound insulation. Besides the standard essential component properties according to building site building acoustical frequency range from 100 Hz to 3150 conditions and tenants‘ needs. Hz, low frequencies (below 100 Hz) can be taken into The maximum significance factor for airborne and impact account “if the user wishes to take a more comprehensive sound insulation values lies at 3.5 % for the overall result and precautionary approach“[14]. if the excellence level is achieved. In case of the entry values their influence is limited to 0.85 % for residential 2.2. BREEAM© GER/AT multi-storey buildings. BREEAM is a protected trademark of the BRE (British 2.3. DGNB©/ÖGNI© Research Establishment) Group. Developed in 1990 as one of the first environmental assessment systems, it has DGNB (Deutsches Gütesiegel für Nachhaltiges Bauen) was become the world’s leading certification system for introduced to the market in 2007 as the sustainability sustainable built environment. This is also due to the fact building assessment system of the German Sustainable that documentation and evidence requirements have Building Council. Since founding, more than 10,000 been broadly adapted to national standards via projects in 27 countries have passed a successful DGNB conformity lists or regional certification bodies have been certification. In more than 50 countries partnerships with established. For Germany and Austria, this task is fulfilled local organisations exist. In the German-speaking area by TÜV SÜD Industrie Service GmbH. ÖGNI – Österreichische Gesellschaft für Nachhaltige For Germany, the following applies: Depending on the Immobilien is responsible for the third party evaluation for degree to which the requirements of DIN 4109-1 [15] are Austrian projects. exceeded or undercut, 1 to 4 points are awarded: 1 point For German projects: 90 points are reserved for airborne corresponds to 1 dB, 4 points to 5 dB exceeding (airborne sound and impact sound insulation of separating building sound insulation values DnT,w + Ctr) or undercutting (impact components, additional 20 points for room acoustics sound insulation values L‘nT,w) of the requirements defined criteria. In total 110 points can be reached, what leads to in the national standard [16]). a maximum of 4.2 % impact of the acoustic criteria in the For Austria, the limit values to be achieved refer to total DGNB scheme [19]. globally defined target values as defined in the BREEAM For Austrian projects (ÖGNI), still the old scheme is in International New Construction Scheme 2016 [17] and force, what leads to an impact of acoustics of 2.3 % in the have not been adapted to higher requirements of the overall result (only airborne and impact sound insulation national building regulations. The highest rating level for max. 1.6 %) [20]. Details to the requirements are listed in airborne sound insulation of separating walls and floors table 2 (airborne sound) and table 3 (impact sound). including spectrum adaption term Ctr is defined as following: DnT,w + Ctr ≥ 53 dB (4 points), the entry level Components Requirement Credit points corresponds to DnT,w + Ctr ≥ 48 dB (1 point). For impact Separating walls DnT,w ≥ 55 dB 5 sound insulation, L'nT,w without spectrum adaption terms DnT,w ≥ 57 dB 10 is used for evaluation: the following performance levels DnT,w ≥ 59 dB 15 are defined for high performance L‘nT,w ≤ 54 dB (4 points) DnT,w ≥ 61 dB 20 and the entry level: L‘nT,w ≤ 59 dB (1 point). Both values Separating floors DnT,w ≥ 55 dB 7,5 exceed the maximum permissible limits for L‘nT,w for DnT,w ≥ 58 dB 15 separating walls/floors with and without openings of the DnT,w ≥ 61 dB 20 Austrian building code [18], which, conversely, means that all projects with Austrian building permission meet the Table 2. ÖGNI airborne sound insulation requirements requirements for the best BREEAM rating for impact and related credits sound insulation if validated via measurements. Dolezal et al.: Acoustic Performance as a Parameter for Ecological and Social Sustainability Assessments 185 AAAA – 2023 – IZOLA - Conference Proceedings Additional criteria are exterior noise (max. 15 points) and building scheme. It is possible to evaluate buildings insulation against noise from HVAC (max. 20 points). according to the following system variants: BNB-B (office buildings), BNB_U (educational buildings) and BNB_L Components Requirement Credit points (laboratory buildings). The BNB system comprises around Separating floors LńT,w ≤ 48 dB 7,5 150 indicators which are grouped to 46 category profiles LńT,w ≤ 43 dB 15 and 11 criteria groups. Therefore, the overall weighting of L´ both indicators - airborne sound as well as impact sound nT,w ≤ 42 dB 20 insulation lies only at 2.7 % (in case of best rating) or 1.35 Table 3. ÖGNI impact sound insulation credits % (entry level). If only one indicator is considered, the mentioned significance factors are halved. 2.4. IBO Ökopass The IBO Ökopass is an Austrian sustainability assessment 3. INTERRELATION BETWEEN SUSTAINABILITY AND system of the first generation which was especially ACOUSTIC PROPERTIES OF BUILDING COMPONENTS designed for residential buildings in urban context. Developed in 2001 by the Austrian Institute for Healthy Environmental impacts of acoustically investigated and Ecological Buildings (IBO), it was applied mainly to building components have been assessed according to the subsidised multi-storey housing projects. The focus lies on standardized, methodologies and results are compared. health- and comfort-related criteria (60% weighting in four main categories), besides the environmental 3.1 Methodology performance of the building [21]. Airborne and impact sound insulation as sub-criteria of The methodology for Life Cycle Assessments (LCA) for the building acoustics section are weighted with a total of building products and building components is roughly 8.6 % in the overall system which is the highest provided in EN 15804 [22] and EN 15978 [23] respectively. significance factor amongst the building assessment With regard to Global Warming Potential (GWP), EN systems examined within this study. The IBO Ökopass 15804 prescribes the baseline model of 100 years of the does not only classify the overall building performance, IPCC based on IPCC 2013. Calculation has been carried out but also the subcriteria in 4 levels: excellent, good, very with software SimaPro, based on ecoinvent v3.8 database good and satisfactory, whereby the minimum [24] for materials as well as for energy datasets. requirement level corresponds to the national building In given research project, emphasis was placed on a code. selected set of indicators, including GWP divided into To meet the Excellent level for airborne sound insulation, GWPfossil and GWPbiogenic. All indicators, required according the following acoustical aspect must be fulfilled by to EN 15804, are calculated, but the focal point of separating walls and floors: D interpretation of results is put on GWP. nT,w + C50-3150 ≥ 63 dB. The spectrum adaption term C50-3150 according to EN 717-1 also includes low frequencies from 50 to 100 Hz. 3.2 Measurement and calculation results for different For impact sound insulation, the IBO Ökopass requires an insulation material in a lightweight wall exceptionally high level for the best rating: the weighted standardised impact sound pressure level L' In an acoustic test facility in Vienna, airborne sound nT,w must be lower than or equal to 35 dB and the total value including insulation property of a lightweight metal stud wall with spectrum adaption terms C the principal structure shown in figure 2 was measured. I and as well as for low frequencies CI,50-2500 have to be fulfilled in the following way: L'nT,w + CI ≤ 40 dB and L'nT,w + CI,50-2500 ≤ 45 dB. 2.5. BNB Assessment System for Sustainable Buildings BNB system is used exclusively for German federal buildings and therefore does not include a residential Fig.2. Structure of tested metal stud wall Dolezal et al.: Acoustic Performance as a Parameter for Ecological and Social Sustainability Assessments 186 AAAA – 2023 – IZOLA - Conference Proceedings The wall was made of 2 separated metal studs 75 mm, Sound measurement discovered a lower sound reduction double planked with gypsum plaster board 12.5 mm and index R for the straw filled wall, but a significantly lower filled with different blow in insulation materials (mineral GWPtotal as well. wool, cellulose fibre, straw and wood fibre). Results were compared acoustically (figure 3) and the influence on GWP has been assessed (figure 4) [8]. 4. CONCLUSIONS A critical review of some of the most important GBLs with focus on acoustic indicators leads to the conclusion that they influence results of GBLs quite differently. Usually indicators follow already standardised descriptors as pointed out above. As pointed out in figure 5, the average impact of the acoustic quality in Sustainable Building Assessment Schemes is around 5.9 %, which is quite a lot, considering the high quantity of different indicators, taken into account in the assessment procedure. Fig.3. Sound reduction index of a metal stud wall with different blow in insulation material Fig.5. Average and maximum possible impact of acoustic aspects on results of SBAS for new buildings The analysed metal stud wall showed lower sound reduction index for blow in insulations from renewable sources. In additional tests, a reduction of the (high) density of the blown in straw led to significantly improved results for the sound insulation. Regarding the environmental aspects of the wall, the carbon content of straw leads to the negative GWPbiogenic and to good results for GWPtotal for the production phase. GWPfossil (GWP from fossil sources) is not significantly lower compared to the other insulation variants. This is a consequence of the little impact on GWPfossil of the insulation material in Fig.4. GWP for the metal stud wall with different general within this structure, since influence on this insulation material - production phase indicator of metal studs and gypsum plaster boards is significantly higher. Dolezal et al.: Acoustic Performance as a Parameter for Ecological and Social Sustainability Assessments 187 AAAA – 2023 – IZOLA - Conference Proceedings 5. ACKNOWLEDGEMENTS Europe. Brussels, 2020. Available at: https://eur- lex.europa.eu/resource.html?uri=cellar:9903b325- 6388-11ea-b735- The authors gratefully acknowledge the financial support 01aa75ed71a1.0017.02/DOC_1&format=PDF of the Austrian Science Fund (FWF) research project I [11.] United Nations: Paris Agreement. 2015. 5503-N Engineered wood composites with enhanced http://unfccc.int/files/essential_background/conven impact sound insulation performance to improve human tion/application/pdf/english_paris_agreement.pdf wellbeing. [12.] ISO 717-1:2020. Acoustics - Rating of sound insulation in buildings and of building elements – Part 1: Airborne sound insulation, International Organization for Standardization, 2020. 6. REFERENCES [13.] ISO 717-2:2020. Acoustics - Rating of sound insulation in buildings and of building elements – [1.] IEA. Global Status Report for Buildings and Part 2: Impact sound insulation, International Construction: Towards a Zero-Emission, Efficient Organization for Standardization, 2020. and Resilient Buildings and Construction Sector. : [14.] Nicholas, D., Shane, D. et al. Level(s) indicator Available at: https://www.iea.org/reports/global- 4.4: Acoustics and protection against noise: User status-report-for-buildings-and-construction-2019 manual: introductory, briefing, instructions and [2.] Ebert, T., Eßig, N., Hauser, G. Zertifizierungssysteme guidance, Publication version 1.1. JRC Technical für Gebäude. Nachhaltigkeit bewerten. Reports, ed. JRC, Unit B.5, 2021. Available at: Internationaler Systemvergleich. Zertifizierung und [15.] DIN 4109-1:2018. Schallschutz im Hochbau – Teil Ökonomie. Institut für internationale Architektur- 1: Mindestanforderungen, Deutsches Institut für Dokumentation GmbH und Co-KG. Redaktion Normung, 2018. DETAIL. München 2010 [16.] BREEAM DE Neubau 2018, Technisches [3.] European Commission (DG ENTR) (ed.): Handbuch, SD BNBDE01, Version 1.1., Stand 06/2022 Standardization Mandate to CEN - M/350 EN, (ed. TÜV SÜD Industrie Service GmbH – NSO BREEAM development of horizontal standardized methods D-A-CH, BRE Global Ltd.) Available at: for the assessment of the integrated environmental https://breeam.de/support/downloads/ performance of buildings. Brussels 2004. [17.] BREEAM AT Neubau 2019, Technisches [4.] EN 16309:2014. Sustainability of construction works Handbuch, SD BNBAT01b, Version 1.1., Stand - Assessment of social performance of buildings - 11/2022 (ed. TÜV SÜD Industrie Service GmbH – NSO Calculation methodology. European Committee for BREEAM D-A-CH, BRE Global Ltd.) Available at: Standardization, 2014. [18.] Österreichisches Institut für Bautechnik: OIB-RL 5 [5.] United Nations. The Sustainable Development Goals Schallschutz, Wien 2019. Report 2022. https://unstats.un.org/sdgs/report [19.] DGNB System. Kriterienkatalog Gebäude [6.] European Commission. The European Green Deal. Neubau, Deutsche Gesellschaft für Nachhaltiges Brussels 2019. Available at: https://eurlex. Bauen. Stuttgart 2023. europa.eu/legal- content/EN/TXT/PDF/?uri=CELEX: [20.] DGNB System. Kriterienkatalog Neubau, ÖGNI 52019DC0640&from=EN TEC1.2 Schallschutz. Österreichische Gesellschaft für [7.] European Commission. Regulation of the European Nachhaltige Immobilienwirtschaft. 2020. Parliament and of the Council laying down [21.] IBO. ÖKOPASS Endbewertung, v.2022. IBO - harmonized conditions for the market of Austrian Institute for Healthy and Ecological construction products. Draft 30.3.2022, Brussels. Buildings. [8.] Neusser, M., Dolezal, F., Wurm, M., Müllner, H., [22.] EN 15804:2022. Sustainability of Construction Bednar, T. Evaluation of the acoustic and Works—Environmental Product Declarations—Core environmental performance of different wall Rules for the Product Category of Construction structures with particular emphasis on straw. Products, European Committee for Standardisation Journal of Building Engineering 66. 2023. Brussels, 2022. https://doi.org/10.1016/j.jobe.2023.105922 [23.] EN 15978:2011. Sustainability of Construction [9.] Teslík, J., Fabian, R., and Hrubá, B. Determination of Works—Environmental Product Declarations— the Airborne Sound Insulation of a Straw Bale Assessment of Environmental Performance of Partition Wall. Civil and Environmental Engineering, Buildings, European Committee for Standardisation vol.13, no.1, 3917, pp.20-29. 2017. Brussels, 2011. https://doi.org/10.1515/cee-2017-0003 [24.] Ecoinvent v2.2 and v3.8, Ecoinvent Database, [10.] European Commission. A new Circular Economy Zurich 2017, 2021. Action Plan for a cleaner and more competitive Dolezal et al.: Acoustic Performance as a Parameter for Ecological and Social Sustainability Assessments 188 IMPROVEMENT OF IMPACT SOUND INSULATION WITH TILE UNDERLAY MATERIALS – IMPACT SOUND INSULATION WITHOUT FLOATING FLOORS Hanna Mária Csiszár1, Gergely Dobszay1, Alajos Miklós Hepke1, Zoltán Hunyadi1, Beáta Mesterházy2, Zsófia Mária Wild1 1 Faculty of Architecture, TU Budapest 2 BME (Budapest University of Tecnology and Economics), Department of Building Constructions, Laboratory of Building Acoustics Abstract: Acoustical experts agree that the impact sound insulation of slabs can be improved with floating floors. The operation and planning methods of such mass-spring systems are well known and they are used widely today. On the basis of Hungarian Central Statistical Office (KSH) in 2021 1 million m2 residential buildings were built in Hungary, wherein in our estimation min. 50.000 m3 estrich screed was built in with the expense of 2 billions HUF (approx. 5,3 millions Euro) and with the CO2 equivalent carbon footprint of approx. 142 290 tons. On the other hand the acoustical quality of existing buildings often can not be improved with floating floors considering the loadbearing capacity and interior height. Building material producers provide different tile underlay materials which can be covered directly, but the use of these materials compared with other solutions is insignificant. The main goal of the research was to assess and quantify the performance and application field of solutions with special underlay materials and to develop possible layer orders without screed. More than 50 measurements were carried out with 10 different underlay materials in the Laboratory of Building Acoustics of BME (Budapest University of Technology and Economics) and in-field. We found that the examined and developed structural solutions can be used to increase the acoustical quality of existing buildings without floating floors and can be applied cautiously in new buildings as well. The base of this research was developed by the students of BME for a Students' Scientific Conference with the guidance of their professors. On the basis of this work we plan to carry out a research project in the future involving building material producers to develop specific solutions and layer orders. Keywords: floating floors, impact sound insulation, resonance frequency, underlay materials, mass-spring system 1. INTRODUCTION improvement, but there is often a demand for hard floor coverings. Floating floors can ensure the acoustical quality Floating floors have been known as a good solution for apart from the top layer of the floor. improvement of impact sound insulation for decades. Although the floating floor is a good solution from the Other floor coverings – e.g. flexible and soft floors – acoustical point of view, it seems “wasteful” considering without floating estrich may have acceptable the environmental protection aspects by reason of high Csiszár – Dobszay – Hepke – Hunyadi – Mesterházy - Wild: Improvement of impact sound insulation with tile underlay materials – impact sound insulation without floating floors 189 AAAA – 2023 – IZOLA - Conference Proceedings cement need. In accordance with European climate policy 2. ACOSUTICAL BACKGROUND it is necessary to develop alternative solutions that are more economical, eco-conscious whilst ensure nearly the The floor coverings may be classified into four groups same quality on the field of impact sound insulation. The from the acoustical point of view: question should be investigated from the point of view of 1. hard floor coverings (e.g. ceramic, stone) glued directly existing buildings as well. Also in accordance with on the slab: these floor coverings can not improve the European climate policy it is necessary to increase impact sound insulation. considerably the rate of refurbishments however the 2. soft floor coverings (e.g. PVC, fitted carpet) glued acoustical quality should be provided or increased if directly to the slab: these floor coverings can improve the possible. These facts justify to seek alternative solutions. impact sound insulation, usually Lw ≤ 30 dB, but the In many existing buildings – typically built with panel improvement under 200 Hz is insignificant. technology, prefabricated or thin reinforced concrete 3. flexible floor coverings (e.g. parquet, laminated slabs – floating floors were not used. The slabs and floor parquet): these floor coverings can improve the impact coverings of these buildings do not fulfil today’s acoustical sound insulation, usually 10 ≤ Lw ≤ 20 dB. requirements that is why they do not provide sufficient 4. floating floors: these floor coverings can improve the acoustical comfort. The original floor coverings – e.g. impact sound insulation, usually Lw ≤ 35 dB. The contact coverings (PVC, fitted carpet), or parquet on characteristics of screed and floating layer can be battens set on slag - are usually changed to hard (ceramic) calculated and planned easily. or up to date flexible (laminated parquet) coverings, Floating floors can be considered as mass-spring system which usually results a worse acoustical quality. from the acoustical point of view. Its crucial elements are The application of floating floor is not a possible solution the mechanical “mass” which can be characterized with its in such buildings due to the limited interior height and specific mass (m’, kg/m2) and the “spring” which can be loadbearing capacity. There is a need for a different characterized with its dynamic stiffness (s’, MN/m3) and solution. The topic is actual as between the 1960-80’s the resistance which may be neglected. The properties of approximately 500.000 panel flats were constructed in the system can be defined with vibration methods. The Hungary. mass-spring system’s significant frequency is the Besides floating floors underlay materials which can be resonance frequency (1) given in [1] and [2]. covered directly are also available, but the use of them is not considerable in Hungary. 1 𝑠′ 𝑓 √ (1) To show the acoustical effect and operation of floor 0 = 2𝜋 𝑚′ coverings containing tile underlay materials we carried out a series of laboratory and in situ measurements. We where investigated the available products and constructed s’ dynamic stiffness (MN/m3) alterative layer orders with the combination of existing m’ specific mass (kg/m2) products considering the questions of stability, rigidity and covering techniques, too. Floor constructions that operate as mass-spring system We assumed that floor coverings with tile underlay improve the impact sound insulation over the resonance materials may have a significant improvement of impact frequency. The goal of the planning process is to decrease sound insulation. We did not assume that they are the value of resonance frequency which can be ensured competitive with floating floors, but hoped that new layer with higher surface mass or softer spring. orders may be used in existing and new buildings, fulfilling We may assume that tile underlay materials with hard the acoustical requirements. We hoped that instead of the coverings operate the same way, where the tile underlay current “wasteful” solutions there are other more material is the “spring” and the hard covering with glue is economical solutions. the “mass”, but it is necessary to investigate that In this research we focused on heavy slabs, thus we question. We considered that the lack of mass will be investigated the effect of developed layer orders with shown especially at lower frequencies. such constructions. Csiszár – Dobszay – Hepke – Hunyadi – Mesterházy - Wild: Improvement of impact sound insulation with tile underlay materials – impact sound insulation without floating floors 190 AAAA – 2023 – IZOLA - Conference Proceedings We examined the concerning previous studies about tile measured in the same combinations as group “B” and “C”. underlay materials. Hongisto et al. [3] investigated layer The total thickness of layer orders was approx. 30-90 mm. orders with 13 different resilient layers and used birch During the preparation and examination of layer orders tree plate as floated layer, the thickness of the layer we realised that there is a need to investigate the orders was 5-20 cm, which exceed the thickness that is questions of stability, rigidity and covering techniques as generally used in residential buildings. Cobos et al. well. Thus parallelly with the acoustical tests we carried proposed tile underlay materials made of industrial waste out static tests related to above mentioned static and used them with floated estrich [4]. Kim K-W et al. problems. This examinations and their results exceed the investigated tile underlay materials having different frame of this paper, hence we do not present these results dynamic stiffness in several layer orders with floated just refer to them if necessary. estrich, too [5]. R. Maderuelo et al. investigated thin (5-10 mm) sound insulation layers made of rubber waste, but the layers on the resilient layer is not detailed [6]. 4. LABORATORY MEASUREMENTS In conclusion the concerning previous studies investigated tile underlay materials and layer orders that differ from In the laboratory we determined the improvement of impact sound insulation of the different layer orders and our concept in some aspects. determined the dynamic stiffness of the resilient layers in addition. The measurements were carried out in the Laboratory of Building Acoustics at the Budapest 3. INVESTIGATED LAYER ORDERS University of Technology and Economics. The measurements were carried out according to ISO 10140 The examined layer orders were classified into four standard series and were evaluated according to ISO 717- groups: 2 standard, the dynamic stiffness measurements were carried out according to EN 29052-1 standard. The type “A” layer orders: The floor coverings (usually 8 mm samples were prepared according to annex H of standard ceramics, or 38 mm concrete paving element) were glued ISO 10140-1. directly on one or more layers of tile underlay material Generally, the acoustical performance of floor coverings (existing product) as recommended by the manufacturers. may be characterized with improvement of impact sound The total thickness of layer orders was approx. 20-30 mm. insulation that can be described with formula (2) given in type “B” (“hybrid”) layer orders: The “hybrid” layer orders [7]. were prepared with multiplication and combination of existing products. The total thickness of layer orders was ∆𝐿 = 𝐿n0−𝐿n (2) approx. 30-40 mm. type “C” layer orders: The “hybrid” layer orders gave the where idea to use ever softer layers from above to downward. In Ln0 is the normalized impact sound pressure level of that combination the upper layer contained the reference floor without the floor covering; reinforcement (glass fibre mesh, polypropylene or Ln is the normalized impact sound pressure level of galvanized steel mesh) to prevent cracks while the lower the reference floor with the floor covering. layers could be more softer, which is optimal for sound insulation. The total thickness of layer orders was approx. The Hungarian national standard MSZ 15601-1:2007 gives 30-60 mm. the basic impact sound insulation requirement between type “D” layer orders: As up to date buildings are usually flats with the quantity of L’nw = 55 dB. supplied with underfloor heating, we wanted to examine The impact sound pressure level is measured between the acoustical effect of the special layers that are used for 100-3150 Hz (optionally 50-5000 Hz), the single number this purpose. Thus we made tests with so called insulated quantity can be determined with a reference curve given studded screed panels. To give a smooth surface for in ISO 717-2 standard. The single number quantity is able gluing the coverings the layers had to be filled in the whole to cover the frequency dependent behaviour of the depth and above them with equalizing mortar which gave structures. an extra weight to the mass zone. These layer orders were The Ln0 curve of the laboratory slab suits well to the Ln,r,0 curve of heavyweight reference floors that is given in ISO Csiszár – Dobszay – Hepke – Hunyadi – Mesterházy - Wild: Improvement of impact sound insulation with tile underlay materials – impact sound insulation without floating floors 191 AAAA – 2023 – IZOLA - Conference Proceedings 717-2 standard. With the difference of the average of is assumed. The calculated resonance frequency of these curves and the curve of reference values for impact sample A.1 (blue curve) was f0=437 Hz (m’= 30 kg/m2, s’= sounds shifted to position L500=55 dB we defined the so 226 MN/m3) which confirms that assumption. The curves called “goal curve”. The “goal curve” shows the required ascend over the resonance frequency with a slope of frequency character of improvement of floor coverings, approximately 10 dB/octave. As the curves nowhere reach without taking into consideration the effect of flanking or exceed the “goal curve”, it is clear that the 20-30 mm transmission which will be discussed later. The reference thin layer orders, which are recommended by the curves and the goal curve are shown in Figure 1. manufacturers have a weak improvement and slabs with them can not fulfil the requirements. Results of type “B” layer orders: The improvement of impact sound insulation of layer order B.3 is shown in Figure 3. Fig.1. The “goal curve” showing the required improvement Fig.3. The improvement of impact sound insulation of layer order type “B” The results of the measurements are discussed in groups. Results of type “A” layer orders: In layer orders “B” one layer mounted heavy polyethylene The improvement of impact sound insulation of layer layer and more layers of rubber-cork were used in orders type “A” are shown in Figure 2. combination. Due to this changes the curves reach and between 200-1600 Hz exceed the “goal curve” although there is a setback around 200 Hz probably due to the resonance frequency. Results of type “C” layer orders: The improvement of impact sound insulation of layer orders type “C” are shown in Figure 4. The first examined layer orders contained just tile underlay materials. As the simultaneously performed failure tests proved the appropriateness of covering zone with reinforcement, we started to use traditional floating layers made of polystyrene, polyethylene or rockwool. As it can be seen in Figure 4 the curves are mostly over the “goal curve”. There is a strong setback at 800 and 1600 Hz Fig.2. The improvement of impact sound insulation of in layer orders in which polypropylene mesh was used as layer orders type “A” reinforcement. The layer orders with rockwool has an outstanding performance. The lower performance of The improvement of these floor coverings is really slight, especially under 400 Hz, where the resonance frequency Csiszár – Dobszay – Hepke – Hunyadi – Mesterházy - Wild: Improvement of impact sound insulation with tile underlay materials – impact sound insulation without floating floors 192 AAAA – 2023 – IZOLA - Conference Proceedings polyethylene layers is probably related to the smaller Hz. The curve of sample D.5 shows a weaker performance thickness of these layer orders. which is probably due to the relatively thin layer order. It is interesting to compare a layer order from group “D” with a layer order from group “A” which is shown in Figure 6. These layer orders differ in one aspect: A.2. does not contain the so called insulated studded screed panel. The beneficial effect of this panel is obvious. Fig.6. Comparison of the improvement of impact sound pressure level of layer orders A.2 and D.1 Fig.4. The improvement of impact sound insulation of 5. IN-SITU MEASUREMENTS layer orders type “C” The in situ measurements were carried out in an existing Results of type “D” layer orders: building built with panel technology and in one of the The improvement of impact sound insulation of layer buildings of the Budapest University of Technology and orders type “D” are shown in Figure 5. Economics. The slabs of panel buildings were usually built with 15-18 cm reinforced concrete and typically soft floor coverings (PVC, fitted carpet) were applied on the top of them which gave a minimal improvement of impact sound insulation. The residents changed the floor coverings typically to ceramic tiles in wet rooms and to laminated parquet or fitted carpet in the rooms. The ceramic tiles and laminated parquet coverings usually result lower acoustical quality than the original coverings. Figures 7 and 8 show the normalized impact sound pressure level of the existing slabs with ceramic tile (Figure 7) and laminated parquet (Figure 8) in the panel building. The single number quantities are L’nw=67 dB and L’nw=61 dB respectively, which exceed significantly the Fig.5. The improvement of impact sound pressure level requirement. of layer orders type “D” As it can be seen in Figure 5 the curves are mostly over the “goal curve”, but there is a significant setback over 1250 Csiszár – Dobszay – Hepke – Hunyadi – Mesterházy - Wild: Improvement of impact sound insulation with tile underlay materials – impact sound insulation without floating floors 193 AAAA – 2023 – IZOLA - Conference Proceedings improvement compared to other layer orders which is probably due to the smaller thickness. Fig.9. The improvement of impact sound pressure in panel building with layer orders A3, B1, D1, D2 Fig.7. Normalized impact sound pressure level in panel building with ceramic tile floor covering, L’n,w=67 dB Existing stairs may be an important application area of tile underlay materials. Today we build heavy stairs with resilient “springs” at the support of stair flights or landings and with isolation between the stair flights, landings and walls as it is described in [8]. Stairs have a monolithic contact in existing buildings thus the noise is transmitted to the rooms through the constructions. These stairs do not fulfil today’s acoustical requirements but the required acoustical isolation can not be provided afterwards. The application of underlay materials may give a solution to these problems. To illustrate the possible effect of layer orders made with underlay materials we carried out in situ measurements in staircases of existing buildings. We determined the difference of impact sound pressure level of existing structures with and without developed layer Fig.8. Normalized impact sound pressure level in panel orders. In this measurements we examined layer orders of building with laminated parquet floor covering, group “A”, the results are shown in Figure 10. L’n,w=61 dB As it was assumed all of the layer orders had an improvement compared with bare constructions, but the To demonstrate the possible improvement of the thicker layer orders with multiple resilient layers showed developed layer orders, we prepared small samples of the a much better performance. following layer orders: A3, B1, D1, D2 and determined the improvement of impact sound insulation compared to the original slab with ceramic tiles. Results are shown in Figure 9. It is obvious that the effect of layer orders which contained insulated studded screed panels (D.1 and D2.) is approximately the same than the effect of layer orders which contained multiple layers of tile underlay material (A.3 – it contained 3×4 mm rubber-cork). B1 layer order – which contained 2×4 mm rubber-cork - shows a weaker Csiszár – Dobszay – Hepke – Hunyadi – Mesterházy - Wild: Improvement of impact sound insulation with tile underlay materials – impact sound insulation without floating floors 194 AAAA – 2023 – IZOLA - Conference Proceedings the rest. With the frequency dependent impact sound pressure levels the layer orders may be optimized. It is interesting to investigate the thickness and the improvement of impact sound insulation of layer orders which can be seen in Figure 12. Fig.10. The improvement of impact sound insulation of stairs of existing buildings with layer orders A.2, A.3, A.5 6. CONCLUSIONS, FURTHER PLANS In the end of the study we determined the single number quantities of the examined layer orders that are shown in Figure 11. Fig.12. The thickness and improvement of impact sound insulation of layer orders It is clear that the really thin layer orders (20-30 mm – group “A”) do not give appropriate improvement. Thicker layer orders (min. 50-60 mm) may reach sufficient improvement. From that point of view group “C” seems a good compromise. The most important results of the project are the following: - 28 layer orders were prepared and measured altogether, more than 50 measurements were carried out which Fig.11. The weighted reduction of impact sound pressure included 38 laboratory measurements and 12 in situ level of layer orders measurements, which contained 8 measurements related to stairs. As we emphasized previously the compliance of a layer - During evaluation of data we made comparative order should be examined together with the slab analyses. Illustrating the frequency related improvement construction. The in situ performance may be determined of coverings we defined a “goal curve” which shows the according to ISO 12354-2. difference between the Ln0 curve of reference floor and We made calculations with the simplified model of ISO the average requirement. With this we could illustrate the 12354-2 and determined that the examined layer orders frequency dependent character of certain solutions and may be applied with heavy slabs which weighted identify the tendencies. normalized impact sound pressure level is between - We found that thin (20-30 mm) layer orders that were Ln,w=64-79 dB. In this project we did not examine the made on the basis of recommendation of manufacturers different slab constructions, but it may be assumed that can not provide such improvement that would realize the the typical weighted normalized impact sound pressure application of these layer orders. level of existing slabs is usually Ln,w=72-85 dB. That means - We found that layer orders made with multiplication and that many of the examined layer order may give a suitable combination of layers may provide significant impact sound solution with most of the existing slab improvement beside economical thickness (50-60 mm). constructions, and may give a significant improvement to Csiszár – Dobszay – Hepke – Hunyadi – Mesterházy - Wild: Improvement of impact sound insulation with tile underlay materials – impact sound insulation without floating floors 195 AAAA – 2023 – IZOLA - Conference Proceedings - We found that Insulated studded screed pane are not [8.] ISO 12354-2:2017 Building acoustics. Estimation of able to replace sound insulating layers, but may be acoustic performance of buildings from the performance of elements. Part 2: Impact sound combined with them for a better quality. insulation between rooms. - We limited the slabs that can be able to fulfil requirements with the examined new layer orders. - We declare that the examined layer orders are able to improve impact sound insulation, acoustical comfort and life quality of existing buildings. In the future we plan to carry out a research project involving building material producers to develop specific solutions and layer orders. Our goal is to specify the acoustical performance, the application area of existing materials and to develop layer orders without floated estrich as a possible alternative for floating floors, considering questions of stability, rigidity and covering techniques. We would like to extend the test to lightweight slab constructions as well. The further goal is to create complex solutions that are suitable to improve the acoustical quality of existing buildings and can be appropriate in new buildings as well. 7. REFERENCES [1.] Reis, F. Az épületakusztika alapjai, Terc, 2003, ISBN: 963 86303 6 1 [2.] P. Nagy J. A hangszigetelés elmélete és gyakorlata, Akadémiai Kiadó, 2004, ISBN: 963 05 8133 7 [3.] Hongisto, V., et al. Acoustic Properties of Commercially Available Thermal Insulators - an Experimental Study, Journal of Building Engineering, vol. 54, 2022. [4.] Cobos, F.J.G and Maderuelo-Sanz, R. Using different waste as resilient layers for impact sound insulation improvement: New alternative to commercial layers?, Building Services Engineering Research and Technology, Volume 43, Issue 4. [5.] Kim, K-W et al. Correlation between dynamic stiffness of resilient materials and heavyweight impact sound reduction level, Building and Environment, Volume 44, Issue 8, 2009. [6.] Maderuelo-Sanz, R. et al. The performance of resilient layers made from recycled rubber fluff for impact noise reduction, Applied Acoustics, Volume 72, Issue 11, 2011. [7.] ISO 140-1:2021 Acoustics. Laboratory measurements of sound insulation of building elements. Part 1: Application rules for specific products. Csiszár – Dobszay – Hepke – Hunyadi – Mesterházy - Wild: Improvement of impact sound insulation with tile underlay materials – impact sound insulation without floating floors 196 Contributed papers Soundscape and sound reproduction techniques 1. The soundscape of university campuses: sound essays on the Polytechnic University of València Alberto Quintana-Gallardo (Centre for Physics Technologies (CTFAMA), Universitat Politècnica de València, Spain) 2. The Accuracy of Dynamic Sound Source Localization in Binaural Audio Systems with Head-Tracking Utilizing Generic and Individual HRTFs Vedran Planinec (Faculty of Electrical Engineering and Computing, University of Zagreb ) 197 THE SOUNDSCAPE OF UNIVERSITY CAMPUSES: SOUND ESSAYS ON THE POLYTECHNIC UNIVERSITY OF VALÈNCIA Alberto Quintana‐Gallardo; Ignacio Guillén‐Guillamón Center for Physics Technologies, Universitat Politècnica de València, València, Spain Abstract: The Polytechnic University of València (UPV) is the biggest polytechnic university in Spain. This university, which was founded in 1971, started as a small Education Institute and now has more than 28000 students in a 620000 m2 university campus and has deep roots in the overall culture of the city of Valencia. This study reflects the acoustic heritage of UPV Campus de Vera and its value as an ever‐changing cultural heritage. The objective is to produce sound essays that capture the ephemeral soundscape of the UPV campus and help the university community reflect on its influence on their day‐to‐day and even in the broader culture of the city. The methodology consists of three distinct phases. The first phase is a bibliographic study of the history of the UPV. This is done by analyzing the architectural projects that defined the university and the published books on the topic. This part is crucial to put the study into context. The second part is recording soundscapes on the campus using ambisonics microphones, sound walks with binaural microphones, and interviews with professors, students, and other university employees. The last part involves editing the audio using a Digital Audio Workstation (DAW). The study results are four sound essays distributed as podcasts on online platforms. After concluding the project, it can be stated that reflecting on the acoustic heritage of university campuses can help students, professors, and other employees create a sense of belonging and ultimately foster their formal and non‐formal educational path. Keywords: Soundscapes; acoustic heritage; sound essays; university campuses 1. INTRODUCTION The idea of soundscapes was first introduduced in the 60s with a groundbreaking initiative called the World Cultural heritage is often understood in visual terms. Soundscape Project. This project, led by the Canadian Preserving the heritage of a city involves maintaining the composer Murray Schafer, paved the way for a new visual integrity, shape and colours, of the architectural generation of acousticians who started to focus on the landmarks. However, the acoustic heritage is commonly uniqueness of sound in each space [1]. forgotten. The auditory experience is essential to the way Soundscapes refer to the acoustic environment of the we perceive spaces. The social context, nature and other surroundings of a particular location [2]. They encompass kinds of sound sources that coexist at a particular time all the sounds that shape the auditory perception of a create soundscapes that, not only tell a story, but also specific place or space. This term goes beyond noise or configure the cultural identity of a place and time. music. It emphasizes the totality of sounds and their Contrary to other forms of heritage, the acoustic heritage interaction with the surrounding environment. The World does not leave a physical mark. The ephemeral nature Soundscape project also introduced the idea of sound soundscapes due to the physical nature of sound has not walks. Sound walks are experiential journeys that involve allowed humanity to know what ancient spaces sounded the listener moving, intentionally or unintentionally, like. However, since the invention of modern‐day through the location. They focus specifically on exploring recording techniques, it has been possible to capture and and engaging with the soundscape of a particular location. register soundscapes. Sound walks can be guided or self‐guided, where 1st Author Surname et al.: Paper title 198 AAAA – 2023 – IZOLA ‐ Conference Proceedings participants actively listen to sounds around them. This innovative architectural trend called Mat‐buiding. Mat study focuses on how to use soundscape recordings to tell buildings are large‐scale, high density structures the story of university campuses in Spain and their organized in a modulated grid [4]. interaction with the cities around them. In the early 70s, during the last years of the dictatorial The design of the building was conceived as a functional regime, several new university campuses were built in system to meet the requirements of the architectural some of the major cities in Spain. These campuses tell an competition, accommodating the needs of all four degree important chapter of the history of Spain during the programs. It was executed in the form of a departmental second half of the 20st century. One of those universities spatial organization, aligning with the prevailing European is the Plytechnic University of Valencia (UPV) [3]. trend of the time, which contrasted with the American approach to campus organization, where each building 1.1. The story of the Polytechnic University of Valencia represented a different school. The building was structured as a grid, with departments strategically The UPV started its activity in 1968 during the latter days positioned, enabling students to tailor their curriculum by of the Franco regime. This university was built in a context moving between these departments. in which the need for qualified professionals was higher The building had a three‐level configuration. The than the rate at which those professionals could be ground floor housed parking and laboratories, while the trained at the learning centers in the country [3]. first floor was dedicated to classrooms. The second floor The UPV was initially named Instituto Politécnico was designated for offices. Notably, the buildings were Superior. It was established in the northern part of the city arranged around an agora, with an east‐west axis in what was at the time cultivation fields. The first building symbolizing power dynamics. On the west side, there was to be built is marked in red on the lower left corner of Fig. the dean's office, representing administrative power, and 1. Designed by Carlos Prat Cambronera y Joaquín on the east side, the central library, symbolizing the power Hernández Martínez, this building hosted at the time the of knowledge [5]. Over the years, the university continued four degrees the Institute offered, Agricultural to expand and incorporate new facilities. In 2002, a Engineering, Industrial Engineering, Civil Engineering and significant change occurred when the concrete platform Architecture [4]. Despite the fact that it was initially was dismantled, and car access was restricted in most planned to be used temporarily until a more permanent parts of the campus. These decisions brought about a building was built, it is in use to this day. The building is profound transformation in the campus's overall structured using a modular grid system with three‐meter appearance. The removal of the platform resulted in intervals. It was designed as a one‐story structure increased isolation between the buildings and the various featuring generously proportioned areas for circulation university departments, diluting the original concept of and branches that serve as classrooms for the various the Mat‐building [5]. Conversely, these decisions also led university degree programs. The most distinctive feature to the creation of more green spaces and a stronger of this design was the inclusion of three diagonally emphasis on integrating nature within the campus. This positioned interior courtyards, strategically placed to included the introduction of various bird species, which offer visual connectivity. The spacious corridors, larger now dominate the acoustic environment in many areas. than usual, reflect a prevailing European trend of the era. This trend emphasized the significance of informal education, recognizing that learning extends beyond the classroom when students engage in conversations and share experiences with each other. During that time, these spaces for social interaction were regarded as equally important as the classrooms themselves. In 1971, a new set of buildings was constructed following the design of the architectural firm L35 Arquitectos and the institute was officially named Universidad Politècnica Fig.1. Map of the Campus de Vera of the Polytechnic de Valencia. The project followed what at the time was an University of Valencia 1st Author Surname et al.: Paper title 199 AAAA – 2023 – IZOLA ‐ Conference Proceedings 1.1. Objectives made available free of charge. Key equipment details are outlined in Table 1. The present study has the following objectives: Because of the dialogical nature inherent in sound essays,  To register the most representative soundscapes of the creative process doesn't adhere to the conventional the present‐day Campus de Vera of UPV. preproduction, production, and postproduction  To find new ways of telling the story of the campus. sequence. Instead, it adopts a recursive approach. The  To articulate the different clusters or groups of various interviews, typically a component of the people that participate in the everyday life of the production phase, have the effect of altering the script university. and sometimes even influencing the preproduction phase  To answer the question, what does the UPV sound itself. The process initiates with the drafting of an initial like? script and the planning of interviews. As each interview is conducted, the script is continually revised and modified based on the interviewee's statements. Once all the 2. MATERIALS AND METHODOLOGY interviews are completed, the script is then finalized, marking the commencement of the postproduction The work is articulated using sound walks, soundscapes, phase. and narrative tools to convey the campus story. Besides that, the work also encompasses using audio‐recorded Description Function Model interviews and narrative fiction. The audio pieces are used Record the in the following way: Ambisonics soundscapes at  Soundscapes: field recordings of the places relevant Zoom VRH‐8 microphone different to each episode using ambisonics or binaural locations microphones, depending on the circumstances. The Binaural Capture the recordings are done following the guidelines of the 3DIO Free Space dummy audio of the international standard ISO/TS 12913‐2:2018. Binaural head sound walks  Sound walks: These walks are carried out by tracing Dynamic Record the a route during which a binaural recording is made, Shure SM58 microphone interviews giving the listener the sensation of walking through Recorder with the sound. Audio Zoom H8 portable microphone  Interviews: designed to involve the university recorder recorder inputs community and to incorporate their subjective Digital impressions. The interviewees are related to the Audio Edit the audio subject matter of each chapter. Adobe Audition Workstation files  Narrations: recorded in the same way as the (DAW) interviews or utilizing a condenser microphone connected to a sound card and processed with a Table 1. Equipment employed to capture the acoustic Digital Audio Workstation (DAW). environment. To recreate sound environments, we will gather the impulsive reactions from selected locations. Impulse responses are used to mimic and simulate sounds in spaces that differ from their original settings through a convolution process. This is accomplished by treating each listening point as a time‐invariant linear system. The various audio components are then blended and edited, organized into thematic chapters for proximity using the Adobe Audition Digital Audio Workstation (DAW). The final work will be uploaded to various digital platforms and 1st Author Surname et al.: Paper title 200 AAAA – 2023 – IZOLA ‐ Conference Proceedings 3. RESULTS boundaries lie. It ponders whether the university's enclosure shields it from the external world or if it isolates Following the initial study, we've devised an the outside from the university. The episode features interdisciplinary podcast framework to present a interviews with individuals who were present during the comprehensive portrayal of the location. This structure early years of the university, accompanied by sound walks mirrors the concept of a dialogue, wherein the narrator along the entrances and boundaries. The primary cedes the floor to other participants, allowing them to objective of this episode is to immerse listeners in the contribute, challenge, and exchange their perspectives on historical context of the university. By understanding this the subject. You can find a detailed outline of this context, students, educators, and other university structure in Table 2. members can gain insights into the origins of many of the current customs and procedures. Section Content Chapter 2. The sound of numbers: This chapter prompts Opening of the Podcast explaining the reflection on what the UPV (University Polytechnic of Intro overarching idea of the sound essays Valencia) actually sounds like. It explores the acoustics of and the project. various campus spaces and the interplay between the An explanation on the content of the natural and technological sounds that coexist within the Hook episode that arises a question that the university's precincts. Sound walks through different dialogue tries to answer areas of the campus will be interwoven with interviews The narrator starts the dialogue with a featuring experts in architectural acoustics, Narrator on opening statement on the topic. This environmental specialists, and researchers in fields such opening part will include facts, data and open as telecommunications and computer science, as well as monologue questions to be answered along the other experts in digital signals. The primary aim of this episode chapter is to foster a closer connection between the After the narrator, several interviews university's community and their surroundings, facilitating Interviews will be concatenated and mixed with a deeper understanding of both the built environment and and speeches soundscapes. the natural elements within the campus. Soundwalks will be recorded in chosen Chapter 3. Neverland: Generational tensions at the UPV. Narrated routes inside the campus. The The university lives in continuous generational renewal, soundwalks soundwalk will include a narration with an eternally young and changing student body that describing some parts of the path. contrasts with the life of the workers. This chapter will Pieces of narrative fiction on the topic deal with the generational tensions in the university and of the episode written and narrated by how its protagonists live it. This chapter aims to bridge the Narrative local writers. The objective is to gap between the different generations that coexist in the fiction approach the topic not only from facts university. Alleviating the generational tension is key to but also from the collective improve the relationship between students and imagination. educators. Understanding each other’s background can improve the learning environment and their mental The narrator gives a closing monologue health. Conclusion encompassing the most important takes Chapter 4. The end of the Polytechnic: This chapter seeks from each intervention. to imagine the future of the university and its possible end. When will the UPV end? There may come a day when Table 2. Structure of the episodes universities are no longer necessary; perhaps it will end when it is no longer needed when technology kills The planned division into chapters aligns with our goal of technology itself. Futures in speculative fiction. The creating a comprehensive portrayal of the campus: purpose of this chapter is to foster critical thinking on Chapter 1. Seeds and limits: This chapter embarks on an what the purpose universities have and what is the overall exploration of the university's origins and its boundaries. purpose of pursuing a university degree. Reflecting on this It delves into the quest to determine where the university's realm commences and where the city's 1st Author Surname et al.: Paper title 201 AAAA – 2023 – IZOLA ‐ Conference Proceedings can help in finding new ways of learning and teaching, as well as to create a more aware university community. 4. REFERENCES [1] H. Boucher and T. Moisey, “An Experiential 3. CONCLUSION Learning of a Philosophy of Music Education Inspired by the Work of Canadian Composer R. Murray Schafer,” Creat. Educ. , vol. 10, no. 10, One of the key objectives of this study was to 2019. emphasize the necessity of reevaluating the approach to [2] A. M. Botella Nicolás, “The soundscape as sound education and recognizing the significance of fostering art,” Cuad. Music. Artes Vis. y Artes Escenicas, vol. engagement with the university campuses. Engagement 15, no. 1, 2020. holds paramount importance, not just for students but [3] R. Castellanos Gómez and D. Domingo Calabiug, also for educators. This rationale guided our decision to “1969: Las Universidades Españolas a Concurso. Bases, Resultados Y Polémicas,” Proy. Progreso, dedicate a chapter to addressing the generational gap in Arquit. , no. 7, pp. 104–121, 2012. the university. It is imperative to narrow this gap or, at the [4] D. Domingo Calabuig, R. Castellanos Gómez, and very least, find avenues for establishing common ground. A. Ábalos Ramos, “The strategies of mat‐building,” With the advent of new tools, we have the capability to Archit. Rev. , no. 1398, pp. 83–91, 2013. motivate both these groups effectively. [5] R. C. Gómez and D. D. Calabuig, “Project and The accessibility of the tools required for producing system: Instituto Politécnico Superior de educational content like this has never been greater, Valencia,” arq Archit. Res. Q. , vol. 20, no. 4, pp. 357–369, Dec. 2016. making initiatives such as this one feasible with relatively modest funding. An essential attribute that renders podcasts and sound‐based content particularly suitable for non‐formal educational projects is their ability to accommodate multitasking. While online educational platforms such as Coursera, EdX, or Khan Academy are invaluable resources, they demand undivided attention. In contrast, podcasts enable listeners to engage while simultaneously carrying out other activities. This characteristic can serve as a solution for individuals with time constraints and helps eliminate barriers to their initial engagement. After this study, the following conclusions can be drawn:  Experimental podcasts can be a viable tool for generating engagement in the university community.  Leveraging audio formats to craft fresh narratives holds significant potential, particularly given the prevalent dominance of our sense of sight. This approach can tap into the power of sound to engage and captivate audiences.  Initiatives like these can serve as valuable supplements to formal education and serve as wellsprings of inspiration for both students and educators. They have the potential to enrich the learning experience and stimulate creative thinking within the academic community. 1st Author Surname et al.: Paper title 202 Accuracy of Dynamic Sound Source Localization in Binaural Audio Systems with Head-Tracking Utilizing Generic and Individual HRTFs Vedran Planinec, Marko Horvat, Kristian Jambrošić, Petar Franček Faculty of Electrical Engineering and Computing, University of Zagreb Abstract: The constant advances in spatial audio technology have the potential to revolutionize the overall audio experience through the integration of binaural systems with head-tracking devices. The hypothesis is that the use of individual Head-Related Transfer Functions (HRTFs) in binaural synthesis benefits the listeners by providing a deeper immersion into the sound scenarios created for virtual and/or augmented reality setups, and, ultimately, a more realistic listening experience. This paper presents the results of a listening experiment in which the listeners were asked to determine the position of a dynamic virtual sound source presented to them binaurally in the horizontal and the median planes, with the goal of assessing how accurately this determination can be done. The experiment was conducted in a control ed listening room environment using a wired head tracker in conjunction with headphones for binaural audio reproduction. The influential factor investigated in the experiment were the HRTFs, which have been varied between a generic HRTF set and the individual sets measured for each listener. The aim is to evaluate if there is any difference in localization accuracy between these two cases and to evaluate its significance through statistical analysis. Keywords: Individual HRTFs; Binaural audio; Head tracking; Listening experiment 1. INTRODUCTION The acquisition of the individual HRTF for a given person must be performed using one of the many available Binaural audio systems are used to create an methods and measurement setups [7]. For this study, the immersive and realistic listening experience that gives the methodology described in [8, 9] was used to determine listener a sense of spatial presence and directionality. individual HRTFs for the subjects who participated in the With the rise of virtual and augmented reality experimental measurements. technologies (VR and AR), binaural audio systems have The purpose of this paper was to conduct experiments to become even more important as they play a crucial role in test the hypothesis that the use of individual HRTF creating a compel ing sensory experience [1]. Binaural increases the accuracy of localization of virtual sound audio systems utilize Head-Related Transfer Functions sources, providing a more realistic and natural experience (HRTF), which are essential for localizing sound in space for users, compared to the situation where generic HRTFs and creating spatial awareness [2]. From gaming and or another person's individual HRTF are used. A novel entertainment [3] to medicine [4] to education and methodology for experiments was implemented that training [5], binaural audio systems and spatial audio are allows the hypothesis to be tested in highly dynamic revolutionizing the way we interact with virtual and virtual scenarios that replicate both real and virtual augmented environments. environments [10]. While generic HRTFs are widely used in spatial audio to create a more accurate VR/AR experience [6], individualized or personal HRTFs can potentially personalize the experience even more. Planinec et al.: Accuracy of Dynamic Sound Source Localization in Binaural Audio Systems with Head-Tracking Utilizing Generic and Individual HRTFs 203 AAAA – 2023 – IZOLA - Conference Proceedings 2. MEASUREMENT METHOD The process of obtaining individual HRTF was conducted in accordance with [8], with the exception that the measurements were taken in an acoustical y isolated smal anechoic chamber (as seen in Fig. 1. ). Fig.2. Horizontal and vertical paper tapes used for determining the azimuth and the elevation of the sound source To avoid negative values for azimuth and elevation, which would be more difficult for the participants to read, the frontal direction was assigned an azimuth and elevation of Fig.1. HRTF measurement in the anechoic chamber 90°, resulting in a range of 0° to 180° for both parameters. Participants were asked to determine the initial and final The binaural sound system developed for the localization azimuth (or elevation) of a virtual sound source moving in accuracy experiment was implemented using a high-the horizontal (or vertical) plane within the selected range quality open-type headphone set, a high-quality external of azimuth (elevation) angles. sound card, a PC, a wired head tracker with a PC The sound stimulus used in this experiment was a single application to send tracking data via the OSC protocol knock on a piece of wood repeated at a rate of 3.53 knocks [11], REAPER v6.75 as the software of choice for the Digital per second. The total duration of the stimulus was Audio Workstation (DAW) [12], and a Sparta Binauralizer randomized and ranged from 8 to 12 seconds. The VST plug-in [13, 14]. direction of arrival of the sound source changed Prior to participating in the experiment, participants were continuously from an initial value to a final value with an provided with an informed consent form and written inconsistent angular velocity (between 4° and 10° per information and instructions about the research and second). experiment. Both the headphones and the head tracker The angular velocity of the virtual sound source was were placed firmly on the participants' heads. To randomized (by no more than 3° per second from the determine the direction in which the sound was arriving, initial velocity) to avoid automatic rotation of the head participants sat on a swivel chair and were free to swivel after adaptation to the constant velocity of the sound their body and head. To localize the sound source and source (in the case of a constant angular velocity). As an determine its azimuth and elevation directions, two paper additional measure to avoid automatic rotation of the bands were stretched symmetrically over a 180° arc head, in some cases the sound stimulus was allowed to around the listening position in the horizontal and vertical have a rotational azimuth in the horizontal plane, i.e., the planes, with the azimuth and elevation angles marked at position at which the virtual source changes its direction 1° resolution (as seen in Fig. 2. ). of motion, i.e., begins to move back toward the initial The radius of the paper arc was 1.5 m, and the precision position. The vertical and horizontal planes were tested of the markings was checked with the digital protractor. independently, and participants were informed which paper band to focus on before each individual Planinec et al.: Accuracy of Dynamic Sound Source Localization in Binaural Audio Systems with Head-Tracking Utilizing Generic and Individual HRTFs 204 AAAA – 2023 – IZOLA - Conference Proceedings measurement. Participants were instructed to verbal y 3. RESULTS indicate the initial direction of the virtual sound source immediately after its localization and the final direction The following section presents the results of the from which they perceived the sound after the sound statistical analyses of the raw data col ected during the stimulus had ended (as shown in Fig. 3. ). experiments and the adjusted raw data. In the statistical analysis, all statistical significance decisions were made at the 0.05 significance level. The statistical analyses were performed in R [16]. The statistical analysis used repeated measures of analysis of variance (ANOVA) [17]. It was performed according to [18]. There were 25 independent participants with multiple measured data. Their results with the same parameters (HRTF, vertical or horizontal plane, starting or ending angle) were averaged as independent data points. Each participant contributed 12 data points for the analysis. Fig.3. A participant with head-mounted binaural system For each of the three HRTFs, there existed perceived elements on a rotating chair, tracking the virtual sound sound source directions represented by their azimuth or stimulus elevation angles, indicating horizontal or vertical starting and ending directions. The obtained data was statistically Prior to the start of the actual test, two test examples analyzed by computing the absolute (non-negative) values were played to each participant to verify the functionality of the deviation from the correct azimuth (or elevation) of the binaural system and to ensure that participants starting and ending direction of the virtual sound source. understood the experimental protocol. In the event of missing data arising from participant In total, each participant listened to 18 test cases and uncertainty regarding the starting or ending angle of the indirectly evaluated the effectiveness of three different virtual sound source, the average absolute deviation for HRTFs. The three HRTFs studied were the participant's the corresponding data point was computed without the own individual HRTF, a generic HRTF obtained from the missing value by utilizing other available deviation data Neumann KU100 dummy head [15], and another person's from the same data point. It is noteworthy that there was individual HRTF obtained from a single participant who always at least one selected angle for each data point for had his individual HRTF measured but did not participate each participant, resulting in no missing data from the in the experiment. The SOFA file containing the generic independent data points for all the participants, which HRTF was the only one with 2702 defined directions, was a crucial factor regarding the proper usage of whereas both individual HRTFs had 1460 defined repeated measures of ANOVA. directions (3° resolution). While a higher number of Given the small number of outliers (less than two outliers directions measured for the generic HRTF could imply a for every 25-point data set), the mean/median imputation potential advantage of the generic HRTF over the technique was employed to adapt outlier values for individual HRTFs in terms of localization accuracy in the ANOVA. Normality testing using Shapiro-Wilk's test was horizontal plane, it is unlikely to have an impact on performed on all data sets, demonstrating that the data is localization accuracy in the vertical plane. normally distributed. Finally, Mauchly's test of sphericity To minimize the possibility of developing a bias, was used to verify that the nul hypothesis of equivalent measurements in the horizontal and vertical planes were variances of differences was fulfil ed. Box and whisker systematical y alternated every three measurements. In plots were generated for the absolute deviation from true addition, the three HRTFs were randomly rearranged after horizontal and vertical starting and ending points, as each segment with three measurements to minimize presented in Fig. 4. . potential confounding effects. Planinec et al.: Accuracy of Dynamic Sound Source Localization in Binaural Audio Systems with Head-Tracking Utilizing Generic and Individual HRTFs 205 AAAA – 2023 – IZOLA - Conference Proceedings Fig.4. Box and whisker plots for the absolute deviation from a true angle as the measure of localization accuracy, depending on the used HRTF, for: (a) the horizontal starting angle, (b) the horizontal ending angle, (c) the vertical starting angle, and (d) the vertical ending angle of the virtual sound source. ANOVA results were calculated as fol ows: • For the horizontal starting angle: F(48,2) = 7.052, p = 0.002. • For the horizontal ending angle: F(48,2) = 3.223, a) p = 0.049. • For the vertical starting angle: F(48,2) = 10.113, p = 0.000216. • For the vertical ending angle: F(48,2) = 4.542, p = 0.016. The results derived from the experiment demonstrate the presence of statistically significant variances across all four parameters that were examined, with respect to the employed HRTF. In order to conduct a more comprehensive evaluation of the performance of each individual HRTF in comparison to other HRTFs, a pairwise comparison was executed. b) The outcomes of the pairwise comparison concerning the horizontal starting angle indicate that there is no statistical y significant difference in absolute deviation from the true angle between one’s own individual HRTF and the individual HRTF of another person (p > 0.05). However, a statistical y significant difference is observed between both of these cases and the case when generic HRTF is used (p < 0.05). For the horizontal ending angle, there was no statistical y significant difference between HRTFs. The results obtained from the pairwise comparison indicate a significantly lower absolute deviation from true c) vertical starting and ending angles when one’s own individual HRTF is used, compared to the remaining two HRTFs (p = 0.007 and p = 0.016). Furthermore, the measurement results obtained from the participants were subjected to analysis to check the accuracy of determination of the direction of movement of the virtual sound source. In cases where the angular value of the sound source increased from the starting angle to the ending angle, it was assigned a value of "1". Conversely, if the angular value decreased, it was assigned a value of "0". This process was carried out for both the programmed values and the measurement results to d) Planinec et al.: Accuracy of Dynamic Sound Source Localization in Binaural Audio Systems with Head-Tracking Utilizing Generic and Individual HRTFs 206 AAAA – 2023 – IZOLA - Conference Proceedings determine the correct direction of movement. The plane when utilizing one's own HRTF in comparison to outcomes of this analysis are presented in Table 1. other HRTFs. Subsequent studies could concentrate on comparing the performance of the individual HRTF measurement method employed in this research with Generic One’s Individu- alternative approaches for measuring individual HRTF. HRTF own al HRTF individu- of al HRTF another 5. REFERENCES person Horizontal 98.67% 98.67% 100% [1.] Middlebrooks, J.C. Sound localization. plane In Handbook of Clinical Neurology; Elsevier: Vertical 44.12% 83.56% 55.07% Amsterdam, The Netherlands, 2015; Volume plane 129, pp. 99–116. [2.] Wightman, F.L.; Kistler, D.J. Headphone Table 1. Correct determination of the direction in which simulation of free-field listening. I: Stimulus synthesis. J. Acoust. Soc. Am. 1989, 85, 858– the sound source moves (in percent) 867. [3.] Broderick, J.; Duggan, J.; Redfern, S. The A Chi-Square (Χ²) test was conducted to test whether the Importance of Spatial Audio in Modern Games scores obtained for correct and incorrect determination of and Virtual Environments. In Proceedings of the direction of movement in the vertical plane carry any the 2018 IEEE Games, Entertainment, Media Conference (GEM), Galway, Ireland, 16–18 statistical y significant difference from the values August 2018; pp. 1–9. expected from purely guessing the direction by random [4.] Johnston, D.; Egermann, H.; Kearney, G. The chance. This was particularly important regarding the Use of Binaural Based Spatial Audio in the generic HRTF since the correct determination percentage Reduction of Auditory Hypersensitivity in Autistic Young People. Int. J. Environ. Res. was slightly below 50%. The results of the Chi-Square test Public Health 2022, 19, 12474. imply that no statistically significant difference exists in [5.] Dede, C.; Jacobson, J.; Richards, J. Introduction: this case, i.e. the determination of direction in the vertical Virtual, Augmented, and Mixed Realities in plane using generic HRTFs is essentially based on guessing. Education. In Virtual, Augmented, and Mixed A binomial test was also conducted, and it confirmed the Realities in Education; Liu, D., Dede, C., Huang, R., Richards, J., Eds.; Springer: results of the Chi-Square test (p=0.3961). The presented Singapore, 2017; pp. 1–16. results demonstrate that using one’s own individual HRTF [6.] Berger, C.C.; Gonzalez-Franco, M.; Tajadura-leads to considerably higher accuracy when determining Jiménez, A.; Florencio, D.; Zhang, Z. Generic the direction in which the sound source moves in the HRTF May be Good Enough in Virtual Reality. Improving Source Localization through vertical plane. Cross-Modal Plasticity. Front. Neurosci. 2018, 12, 21. [7.] Li, S.; Peissig, J. Measurement of Head- 4. CONCLUSION Related Transfer Functions: A Review. Appl. Sci. 2020, 10, 5014. [8.] Reijniers, J.; Partoens, B.; Peremans, H. DIY In summary, this study presents a novel Measurement of Your Personal HRTF at methodology for evaluating the localization accuracy in Home: Low-Cost, Fast and Validated. binaural audio systems, with a particular emphasis on Convention e-Brief 399. In Proceedings of the identifying potential enhancements through the use of an 143rd International AES Convention, New York, individual's own HRTF as opposed to a generic HRTF or NY, USA, 18–21 October 2017. [9.] Reijniers, J.; Partoens, B.; Steckel, J.; Peremans, that of another individual. While the findings indicate a H. HRTF Measurement by Means of statistical y significant improvement in accuracy for an Unsupervised Head Movements with Respect individual's own HRTF in the vertical plane, they also to a Single Fixed Speaker. IEEE suggest the absence of universal statistical y significant Access 2020, 8, 92287–92300. [10.] Planinec, V.; Reijniers, J.; Horvat, M.; improvements in localization accuracy in the horizontal Peremans, H.; Jambrošić, K. The Accuracy of Planinec et al.: Accuracy of Dynamic Sound Source Localization in Binaural Audio Systems with Head-Tracking Utilizing Generic and Individual HRTFs 207 AAAA – 2023 – IZOLA - Conference Proceedings Dynamic Sound Source Localization and [14.] McCormack, L.; Politis, A. Spatial Audio Recognition Ability of Individual Head- Real-Time Applications. Available Related Transfer Functions in Binaural Audio online: http://research.spa.aalto.fi/projects/sp Systems with Head Tracking. Appl. arta_vsts/ (accessed on 29 August 2023). Sci. 2023, 13, 5254. [15.] Neumann KU100 Operating Instructions. [11.] Wright, M.; Freed, A. Open Sound Control: Technical Report. Available A New Protocol for Communicating with online: https://www.manualslib.com/manual/1 Sound Synthesizers. In Proceedings of the 10720/Neumann-Berlin-Dummy-Head-Ku- International Computer Music Conference 100.html (accessed on 29 August 2023). (ICMC), Thessaloniki, Greece, 25–30 September [16.] R Core Team. R: A Language and 1997; pp. 101–104. Environment for Statistical Computing; R [12.] COCKOS Inc. Reaper: Digital Audio Foundation for Statistical Computing. Vienna, Workstation. Rosendale, NY, USA, 2023. Austria, 2021; Available online: https://www.R- Available project.org/ (accessed on 29 August 2023). online: https://www.reaper.fm (accessed on 29 [17.] Park, E.; Cho, M.; Ki, C.S. Correct use of August 2023). repeated measures analysis of [13.] McCormack, L.; Politis, A. SPARTA & variance. Korean J. Lab. Med. 2009, 29, 1–9. COMPASS: Real-Time Implementations of [18.] Finnstats. Repeated Measures of ANOVA Linear and Parametric Spatial Audio in R Complete Tutorial. Available Reproduction and Processing Methods. In online: https://finnstats.com/index.php/2021/0 Proceedings of the AES International Conference 4/06/repeated-measures-of-anova-in- on Immersive and Interactive Audio, York, UK, r/ (accessed on 29 August 2023). 27–29 March 2019; p. 111 Planinec et al.: Accuracy of Dynamic Sound Source Localization in Binaural Audio Systems with Head-Tracking Utilizing Generic and Individual HRTFs 208 Contributed papers Noise and vibrations 1. Ilegal use of firecrackers and its consequences - case study of human rights violation at Slovenian courts – part i: legislation and court proceedings Ferdinand Deželak (Retired researcher) 2. Ilegal use of firecrackers and its consequences - case study of human rights violation at Slovenian courts – part ii: physical background Ferdinand Deželak (Retired researcher) 3. Low frequency noise measurement in the passenger cabin Samo Beguš (University of Ljubljana, Faculty of Electrical Engineering) 4. Measurement and Characterization of Control Valves Noise Egon Susič (Danfoss Trata d.o.o.) 209 Ilegal Use of Firecrackers and its Consequences - Case Study of Human Rights Violation at Slovenian Courts – Part I: Legislation and Court Proceedings Ferdinand Deželak Retired researcher Abstract instance, are regularly entrusted to the courts, where it is In first part of this paper a human rights violation at some expected a right and fair solution to be found. Often Slovenian courts is briefly highlighted. This violation has however, successful police actions are entirely revoked as been firmly proved not just through a judicial proofs, but a result of a corruption and erroneous court decisions. also by a full physical analysis. Apart of the court This paper describes one of such cases which was quite proceedings some important physical facts are described, unprofessional and of a highly corruptive nature. It which were quite misunderstood and even intentionally started in the District Court of Ljubljana (Slovenia). Even neglected at more Slovenian courts. The corresponding higher Slovenian courts were not able to solve and explain legislation which was misinterpreted or even ignored is it till now. shortly presented as well. Such ignorance resulted in quite In this article, a presentation of bad work practices at some contradictory and illegal judgments. courts is analyzed. Situations where some Slovenian courts Some years ago, one attacker threw an explosive device at select bad judges and experts witnesses without an the victim, which consequently suffered a permanent appropriate knowledge and experience are not rare. hearing impairment. Although reporting this attack to the Consequently, such expert testimonies can be quite wrong, Police, further criminal and lawsuit proceedings at the including those in the field of acoustics and audiometry. In different courts in Ljubljana against the attacker were this paper one of such cases which is related to a set of corrupted, unprofessional and completely biased by some other poorly guided procedures at these courts, violating judges. law, human rights and dignity, is briefly highlighted. Lawsuit proceedings were started by one corrupted judge, It all started when one perpetrator (the attacker J.L.) threw who was later found guilty of corruption and dismissed a powerful firecracker, which exploded approximately 4 to from judicial service. The next judge then appointed one 5 meters from the right ear of the plaintiff (victim F.D.). This retired court expert witness in order to solve some loud explosion resulted in his hearing impairment, so the audiometric and acoustic problems concerning a victim’s prosecution was submitted at the District Court against the hearing loss as a result of this explosion. However, he was attacker. However, the judges at this court were not not familiar with basic facts of acoustics and was not able interested in finding a fair and legal solution as they fully to do that job correctly. sympathized with the attacker. Therefore, the trial of this court was far from being fair and professional. Keywords: audiometry, corruption, criminal proceeding, Apart of this, the first judge F.K., who started litigating expert witness testimony, firecracker noise, hearing process of this case, was generally not of high moral damage, high impulse noise, human rights violation, character and has been engaged in some activities which litigation, peak sound pressure level. were quite incompatible with his position as an independent and impartial judge of a full-time office and with his general judicial function. Although he was 1. INTRODUCTION suspended after four years of work on this case, another judge of this court, A.B., continued to apply this bad practice, especially in the field of acoustics and audiology, The use of fireworks and firecrackers is strictly limited in which were crucial elements of this case. So, the second most of the European countries. Unfortunately, some judge appointed an expert witness A.G., (who was even people still violate these legislative requirements. In many not a member of the court experts society) in order to cases, police actions mitigate these problems by taking solve professional questions concerning this explosion appropriate measures, especially of a preventive and its consequences to the plaintiffs hearing impairment. character. On the other hand, some problematic cases, However, this expert was not only in a biased position but where violence and personal injuries are involved for Deželak: Ilegal use of firecrackers, Part 1 210 AAAA – 2023 – IZOLA - Conference Proceedings was also unfamiliar with fundamental acoustic and important things, the judge completely refused to audiometric facts, so he has even not been able to provide consider these obligations of court, saying directly that the correct answers to the plaintiff questions and to the she will decide what should be neglected and what not. court. In this way the expert witness has given false and Although the attacker caused permanent hearing damage fraudulent testimony. Most of his methods were wrong or to the plaintiff by its criminal action, he has been unreliable, so even the expert himself was unable to test completely acquitted; even despite this obligation and the and explain them properly. This affected the error rate, so established judicial practice [2], which in such cases it was quite impossible for someone else to reproduce the anticipates for high compensation and criminal liability. process. On the other hand, the new judge permanently However, even the prosecutor herself was not interested intervened and discussed about the professional about seriousness and consequences of the attacker questions, although she did not understand them. In this infringements. way she proclaimed almost all important acoustic and The aggressor who caused permanent health damage to a medical questions as irrelevant. When such a situation victim, by throwing an explosive device was thus does occur, it would certainly place the court and law firm completely released of any responsibility, and all costs of in jeopardy. It comes as no surprise then that many these criminal proceedings were covered from the Slovenian citizens do not trust to such a judicial system. The author primarily focused on the incorrect position of budget, including the costs of his attorney employed by this expert witness and the misleading aspects of his the former Minister of Justice M.K. On the other hand, the testimony, pointing further to bad practice in the District injured victim has been sentenced to cover all the costs by and other courts of higher instances, resulting in quite himself. Moreover, in contrast of the attacker the victim wrong conclusions and judgements. was even sentenced to fully pay the court taxes. It is more than obvious that the District judge M.L.B. 2. EVENT DESCRIPTION AND COURT PROCESSING systematically and consciously violated the law and the On April 10th 2012, the perpetrator threw an explosive human rights of the victim, just in order to fully protect device (a stronger firecracker) at the victim without any the interests of the attacker J.L. In her judgment no. VI K warning. In this attack the plaintiff suffered a permanent 38124/2014 [3] she completely relieved the attacker of hearing impairment, among other things. Such violation any obligations and responsibilities. She did this without include both crimes and civil wrongs. The plaintiff, any appropriate explanation, although it has been proved reported this attack immediately to the Police, which that the attacker intentionally threw an explosive device completed its job. Unfortunately further criminal proceedings at the District court in Ljubljana against the at the victim, causing him permanent injuries. It is not attacker were corrupted and completely biased, starting strange that such misconduct of judge, like a suppressing with the District judge at criminal department M.L.B. The evidence from the victim, encouraging deceit from former Minister of Justice M.K. namely intervened and witnesses and prosecutorial bluffing (threats and protected the attacker with his undue influence and intimidation) further encouraged the attacker, so his managed to reject the existing proofs and criminal violent actions against the victim not only continued, but proceedings against the attacker in this court. even escalated. 2.1. Criminal Trial Dealing with the Attacker Criminal Act Since the same beginning the District judge M.L.B., dealing 2.2. Start of Lawsuit Proceeding with a Corrupted Judge with the attacker criminal liability, constantly tried to persuade a plaintiff to resign from further prosecution. There is no doubt that such trial of the attacker was quite When not successful she tried to divert attention and unlawful and inadmissible. The plaintiff (victim) therefore began harassing and intimidate the plaintiff with also claimed for damages against the perpetrator J.L. at completely irrelevant questions, like whether he had the District Court in Ljubljana for the fear suffered and written some insulting letters to one notary and which permanent hearing impairment caused by his explosive has nothing to do with the present case (except perhaps device. Unfortunately, this civil procedure has been even the judges connections with this notary). On the other more biased and taken over by another corrupted judge hand, she deliberately ignored much more important facts, such as attacker obligations toward the victim. The of the same court, F.K., who has later been found guilty of Ljubljana District Court namely proposed itself and corruption and dismissed from judicial service. He has confirmed earlier another criminal trial IK 962/2004 [1], been sentenced due to acting in ways that are considered about the attacker obligations to avoid any conflict with unethical and since he violated his judge's obligations of the plaintiff. Despite the plaintiffs warnings about these impartial conduct several times. Deželak: Ilegal use of firecrackers, Part 1 1st Author Surname et al.: Paper title 211 AAAA – 2023 – IZOLA - Conference Proceedings The explosion, which took place due to a perpetrator professional standard for the field of interest. Since the attack, has been unpredicted and unexpected, so the plaintiff, himself is an expert in this field as well, he has plaintiff’s ears were totally unprotected and were also been able to professionally observe and critically exposed to a high-impulse sound pressure level exceeding evaluate this expert's work. The plaintiff completed Ph.D. 150 dBC. Independent audiometric tests showed that this in the field of impulse noise [7] and also acted as a court resulted in hearing impairment of the plaintiff. However expert witness himself in some similar tasks. As to the the first judge F.K., who started processing this civil case, position regarding expertise of witness, there were no ignored these audiometric tests completely together with official trial for this to be examined, although such bad the most valid evidence. After a five years his employment evidence is normally be ruled inadmissible. Instead, the at the District Court was terminated due to several court judge A.B., a former minister M.K. and the expert A.G. scandals coming into publicity and which the court was simply took a position of their power and immunity, based unable to conceal any more, so the judge F.K. was apparently on the principle "la loi c/est moi". They simply consequently suspended. understand their immunity as something quite absolute. It is quite obvious that some judges of the District Court in Ljubljana, under the pressure of the former Minister of 2.3. Replacement of a Judge and Appointment of an Justice, did not condemn the planned attack on the Unreliable Expert Witness plaintiff, with a dangerous explosive device. Even more, The case was then handed over to a new judge, A.B., who they even approved this attack on the plaintiff's body, was also immediately pressured by the former Minister of despite quite opposite international and slovenian law Justice. This new judge made it clear at the outset already, practice. A former Minister of Justice M.K. namely that the plaintiff has no chances of succeeding in this case, intervened again in the attacker's favor. In this way, the despite the indisputably proven guilt of the perpetrator case completely deviated from the established practice of (attacker). The former Minister of Justice and the attorney law and norms, just in order to protect the attacker of the attacker even suggested that the plaintiff should be interests for any price. All the plaintiff's appeals were in fined with a high penalty costs, which the new judge A.B. vain. The former Minister of Justice with his political immediately took into consideration and in the end influence further managed to convince the Higher Court actually fined the plaintiff with 5.600 EUR[4, 34, 35, 36]. in favor of the attacker as well. Indirectly, the Supreme and Constitutional Court were under his pressure as well. So, this replacement did not improve the situation in no In this way the case acquired a completely new political way. A new judge A.B. adopted a similar corruptive and corruptive dimensions. practice. She decided to appoint an unreliable expert It is undisputed, and even the attacker himself did not witness in order to resolve some professional questions deny the fact, that he deliberately threw an explosive on acoustics and audiology and to assess the seriousness device on the plaintiff, causing permanent health of this explosion. Apart from her biased position, she damages to him. It was thus proved beyond a reasonable selected the expert witness with an unprofessional doubt that the perpetrator committed the crimes knowledge and experience. It is imperative of course, that charged, which consequently caused damage to the judge must choose their experts wisely, but she did not. plaintiff. Additionally, this has been confirmed by the This further resulted in a quite unprofessional work by this police investigations as well. In this way, the courts expert witness [5], including his poorly planned intentionally ignored all important evidences and proofs. measurements, erroneous calculations and quite All court procedures were conducted quite in contrast to incorrect conclusions. Despite additional cross the proofs and established case law [2], which in such examination [6], he was unable to give correct and logical cases anticipates for high compensation and criminal answers to the most important questions. liability. All previous court obligations of the attacker 2.4. Wrong Expert Witness Working Methods and their about settlement at the District Court in Ljubljana [1], Uncritical Confirmation by the Court where it was required that the attacker must make any So this expert witness testimony was as expected, effort to avoid conflicts with the plaintiff, were designed in advance, heavily biased and completely intentionaly neglected as well. wrong. Expert witness testimony has not been based on a In its judgment [8], even the Higher Court uncritically Deželak: Ilegal use of firecrackers, Part 1 1st Author Surname et al.: Paper title 212 AAAA – 2023 – IZOLA - Conference Proceedings accepted the very similiar judgement as the District Court expert witness A.G. did before (see annex A). It further in Ljubljana did, confirming the unprofessional testimony unfoundedly accused the plaintiff of not having of A.G. On the other hand, although the slovenian courts knowledge about hearing, and then draws the final themselves do not have any knowledge in acoustical field erroneous conclusion that the firecracker's explosion they nevertheless interfere deeply into professional issues could not have caused any acoustic trauma at all, relying of acoustics, making new additional cardinal errors. solely on the erroneous conclusion of the expert witness In a quite similar way the Supreme [9] and Constitutional A.G. again, without any justification. Court of Slovenia [10, 11] confirmed that the plaintiff was The plaintiff repeatedly emphasized the most important not entitled to any compensation for the deliberately issue of the difference between 120 dB (A, imp), as produced explosion, causing consequently a permanent considered by the European legislation on firecrackers health damage to him. Such activities of the four slovenian and 130 dB to which the expert erroneously refers as limit courts not only supported intentionally caused health value for hearing impairment (annex A). Due to the bad impairments, humiliation, material damage to the professional knowledge of expert witness A.G. and plaintiff, his human rights and dignity violation. There is permanent influence of the former minister M.K., this also the reputation of these courts in question, which, by most important issue remained unsolved by all courts; uncritically confirming unprofessional claims and although the plaintiff had explained it several times. erroneous conclusions of expert witness A.G., were quite During the cross examination [6], the expert replied in embarrassed. It is probably not necessary to emphasize confusion that we have a relative decibel and an absolute that by confirming such a corrupted judgment, it has been decibel and that he know these two kinds of decibels from fully accepted and transferred into established case law. an audiological point of view. Despite repeated trials, the The District Court even fined the plaintiff (a victim) expert avoided to explain to this most important question additionally, with an unusually high penalty of over 5.600 about decibel units, which he used incorrectly. To purely EUR [4, 34, 35, 36]. The Higher Court confirmed this in concrete questions, he merely replied in confusion that order to compensate interventions of former minister "he uses a level of 130 to 140 decibels, which can already M.K. [8]. This encouraged him, so after that he even raised lead to health damage." When asked again what decibels that amount [34, 35, 36]. This can be interpreted of he was talking about, the expert becomes upset that he course, that since then the attack with the explosive will not discuss about such questions and the judge A.B. devices should to be considered completely legal. Even fully supported him to refuse this the most important more, in the case of criminal or lawsuit proceedings, the explanation. The use of such incorrect metrics by the attacker can even be financially stimulated and rewarded, expert resulted in a difference of 26 decibels, which while the victim of the attack can expect to be punished. means an error by factor of 400. In this way the expert got In its answer, the Higher Court in Judgment [8] namely a completely wrong results and thus also the conclusion, only routinely repeats the claims of the District Court, that which he was unable to explain. For the sake of the explosion of a firecracker presents no danger at all and transparency, the plaintiff made a comparison of EU that there is no possibility for any hearing damage, Directive [12] and what the expert asserted. Namely, the claiming further that this has been proved by objectively requirement of this European Directive to which the measurable data on blast intensity of the expert witness expert actually refers, demands that the maximum noise A.G. ?! First of all, the court nowhere explained what it level must not exceed 120 dB (A, imp) at the safety understands by "objectively measurable data on blast distance. intensity", since the expert witness had not perform any The difference between the two quotes is obvious. While blast intensity measurements and analysis at all. Even the European directive [12] prescribes a limit of 120 dB (A, more, the expert witness even refused to explain, what he imp), the expert witness A.G. considered only the first part understand under the term blast intensity. The court of this specification, that is 120 dB. He then compares this further reiterates the expert cardinal error that the directly with the hearing protection directive [13], which intensity of the sound due to the firecracker explosion is uses dB (C, peak) as a metrics. Such processing of expert 124.08 dB at a distance of 5 meters (no specification of witness is quite wrong of course (see Annex A). Even decibels were stated). It did not explain which decibels it more, he made some additional mistakes as described has in mind, repeating the same cardinal error as the later. Interestingly, even the Higher and Supreme judges Deželak: Ilegal use of firecrackers, Part 1 1st Author Surname et al.: Paper title 213 AAAA – 2023 – IZOLA - Conference Proceedings stubbornly ignored this most important fact, although was heavily violated. He was not entitled to a fair and they were strictly and explicitly warned about it. Due to public hearing within a reasonable time by an their missing knowledge the judges further confused the independent and impartial tribunal established by law term “metrics” with “matrix,” which of course denotes [16]. something quite different and with which these judges If a witness's testimony at hearing contradicts his earlier again merely ashamed themselves. statements (as happened here as well), one or both In addition, the expert witness also violated Regulation parties might bring up these statement to impeach his 679 EU 2016 on the protection of personal data and on testimony. The control and understanding of these the free movement of such data [14]. The expert witness processes is vital to producing verifiable and reproducible A.G. namely disclosed many sensitive plaintiff's personal conclusions. However, no one of the factors that may be data, although they were not necessary for his testimony considered in determining whether the methodology at all. Even the expert witness himself later admitted that used is valid, were used. No checks, like whether the he did not need nor used numerical data on the plaintiff's theory or technique in question can be and has been blood pressure, cholesterol, triglycerides and the like. tested nor its known or potential error rate were made, Nevertheless he disclosed all these strictly confident despite many plaintiffs objections in this direction. numerical values to the opposite party without the The courts finally took a position that nothing was wrong plaintiff's consent. He did not need these data for expert with such processing. They even claimed that what testimony at all and, of course, he did not use them, as happened during the hearing loss event due to firecracker they have nothing to do with hearing impairment. If they explosion is quiet irrelevant. In this way, they completely had, the courts would have to take them into account, but and intentionally ignored the most important facts; of course, they didn’t. The plaintiff also repeatedly namely that the plaintiff has been ambushed by the pointed out these facts to the court. However, the judge attacker with an explosive device. They further ignored A.B. merely responded vaguely to these warnings that she that sound pressure level at his right ear exceeded 150 dB is not interested about the plaintiff's personal data. (C, peak), which is more than 15 dB (C, peak), or more than Interestingly, during the cross-examination, the District 30 times higher than allowed for the hearing conservation judge A.B. persistently blocked plaintiff's questions by purposes by EU normatives [13]. On the other hand, it has emphasizing that expert was dealing with “audiological been professionally proved that the plaintiff suffered decibels”. During the cross-examining expert witness the hearing impairment as a result of the explosion caused by judge A.B. find herself wading into technical areas where a deliberately thrown firecracker close to his right ear. she was far inferior. She was not prepared to such The Higher Court relied his decision on the short and situation before stepping into the courtroom. Despite this, erroneous arguments of the District Court as well as on a she interfered with most professional questions, although wrong expert witness testimony of A.G., which were she has not been able to explain their definitions and completely erroneous [5]. At the same time, it clearly meanings, or what she was actually talking about, till now. confirms that it did not consider the plaintiff's At the plaintiffs request she simply replied, that this is not professionally substantiated remarks, making fun of his her duty [15]. health, human rights and dignity. In the middle of cross examination she interrupted the The Supreme [9] and Constitutional courts [10, 11] also plaintiff, not allowing him any further question and closed avoided to deal with these most important questions and the process of hearing without finishing necessary just confirmed what the District and Higher court decided, investigation. Here the validity of an expert's testimony without any further explanations. was also challenged due to the methodology used to form Apart of criminal and lawsuit legislation many Directives, his testimony. However, the judge A.B. disabled no laws and standards were violated. Such infringement impeachment, neither allowed any further questions. present a serious violation of human right and dignity of Even more, she blocked any plaintiffs question requiring the plaintiff. the clarification of such experts contradictions and Even more, the executive department of the District Court encouraging him simultaneously not to answer to any [35] issued an order specifying a double amount of perpetrator costs to be paid for the same litigation and further questions. Even higher courts in Ljubljana fined the plaintiff for the additional sum of EUR 1,600.00, supported this behaviour so the plaintiffs right to fair trial i.e. a total of more than EUR 5,600.00. So far, the court did Deželak: Ilegal use of firecrackers, Part 1 1st Author Surname et al.: Paper title 214 AAAA – 2023 – IZOLA - Conference Proceedings not reveal what the background of this failure was: the way. deliberate attacker deception by the court help, or simply an extremely negligence of the court again [35], although their terrible mistakes were confirmed as well [36]. 4.2. Noise Protection and Health Directive Directive 2003/10/EC [13] prescribes limit values of 3. THE ROLE OF A EXPERT WITNESS impulse noise, to which the human ears – usually at workplaces can be exposed. Generally, expert witness must have a reliable knowledge and be experienced in a entrusted court case. His duty is to interpret all important facts and to make a completely 4.3. International Convention for the Protection of impartial and independent testimony relevant to the Human Rights and Fundamental Freedoms issues in the action. Expert witness, on the other hand, According to this Convention everyone has the right for must prove the court the facts about his expertise in order fair trial. No one shall be subjected to inhuman or to convince and to believe in the case at hand. Expert degrading treatment. Everyone is entitled to a fair and witness is called to have an objective testimony in court, public hearing by an impartial tribunal. Everyone whose completely relying on law and his experiences in the case rights are violated shall have an effective remedy before a under consideration. His presence is necessary in order to national authority not withstanding that the violation has explain complicated scientific issues; on the other hand, been committed by persons acting in an official capacity. he can not influence the jury or judge with fervor. If the However, all these articles of this Convention were expert witness violates such of his sworn duties, his violated by different Slovenian courts. testimony should not be considered as admissible [24]. When the expert witness lies, when he willfully fails to take into account alI the circumstances involved in the 4.4. Audiometric Standards case, or when his medical conclusions are outside the In order to obtain a reliable measure of hearing ability and realm of accepted scientific method and practice, it can be its loss, many factors must be considered. For this purpose considered as an unethical and unlawful testimony. some tests are usually performed in specially designed audiometric test rooms. One of the most important requirements, in order to avoid masking the test signal by 4. LEGISLATIVE REQUIREMENTS ambient noise in such a test room, is that the levels of the ambient noise shall not exceed certain values. The 4.1. Fireworks Act and EU Fireworks Directive standards ISO 8253 [18,19] provide threshold levels, which indicate the maximum ambient sound pressure First of all, the use of fireworks and firecrackers is strictly levels which are still permissible when other minimum limited in most of the European countries. According to hearing threshold levels are to be measured. lt sets out Slovenian legislation, the use of certain categories of procedures for determining hearing threshold levels by fireworks is permitted, but only between December 26th pure tone air conduction and bone conduction and January 2nd and even then only at appropriate audiometry. locations [17]. Outside this time interval and these Ambient sound pressure levels in an audiometric test locations their use is strictly forbidden. The penalties for room shall not exceed the values specified in table 2 of the such offenses (even when no other accidents are involved) standard ISO 8253 – 2 [19]. The test subject and the tester range from four hundred to 1.200 EUR. Despite this, the shall be neither disturbed nor distracted by non-related attacker has not been fined at all, even more, he was events nor by people in the vicinity. rewarded. For this reason, permissible ambient noise for threshold According to EU Directive 2013/29/ EU [12], fireworks are determinations in an audiometric test room shall not categorized according to their type of use, or their exceed certain values in order to avoid masking the test purpose and level of hazard, including their noise level, tones. These values are specified as maximum permissible into four categories F1 to F4. They are intended only for ambient sound pressure levels in one-third octave bands outdoor use and minimum safety distance. In this for the lowest hearing threshold level. So, to satisfy Directive these distances and permitted noise levels are ambient noise requirements, it will be necessary in many specified, which the expert witness used in a quite wrong Deželak: Ilegal use of firecrackers, Part 1 1st Author Surname et al.: Paper title 215 AAAA – 2023 – IZOLA - Conference Proceedings circumstances to use a sound-isolating booth. [13] Directive 2003/10/EC on the minimum health and safety requirements regarding the exposure of workers to the risks arising from Furthermore, a calibration interval of not more than one physical agents (noise). year is recommended. [14] Regulation 679 EU 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data and repealing Directive 95/46/EC. 4.5. Personal Data Protection Directive [15] A. Bedene, District Court in Ljubljana, II P 1669/2015, The refusal to explain false allegations during the expert cross-examination, March 6th In EU personal information from individuals can be 2020. collected and processed in accordance with General Data [16] European Convention on Human Rights, European Court of Human Rights Council of Europe F-67075 Strasbourg cedex. Protection Regulation [14]. [17] Explosives and Pyrotechnic Articles Act; Zakon o eksplozivih in Expert witness can collect such sensitive personal data for pirotehničnih izdelkih, O.J. RS, 35/2008. [18] EN ISO 8253 Acoustics- Audiometric test methods- Part 1: Basic pure specified, explicit and legitimate reasons. According to tone air and bone conduction threshold audiometry. this Regulation, the health-related data is considered as [19] EN ISO 8253 Acoustics- Audiometric test methods- Part 2: Sound sensitive personal data and is subject to specific field audiometry with pure tone and narrow-band test signals. [20] P. Teague, J. Conomos and V. Alexandrou »Overview of processing conditions. Developments in the Description and Assessment of High Intensity Personal data must therefore be adequate, relevant Impulse Noise Exposure«, Proceedings of Acoustics 2016, Brisbane, and limited to what is necessary in relation to the Australia, 9-11 November 2016. [21] G.Richard Price: »Impulse noise hazard: From theoretical purposes for which they are collected and processed (data understanding to engineering solutions«, Noise Control Engineering minimisation). Journal 60 (3), pp. 301-312, May - June 2012. [22] G. R. Price: »Predicting Mechanical Damage to the Organ of Corti«, For this reason, the expert witness can use plaintiffs Hearing Research, Vol 226, pp. 5 – 13, April 2007. sensitive personal data, but only in a way that is adequate, [23] Non – binding guide to good practice for the application of directive 2003/10/EC Noise at Work; Chapter 7: Hearing damage and health relevant and not excessive. He is obliged to process surveillance, European Commission Directorate-General for personal data only where it is necessary to do so and for Employment, Social Affairs and Equal Opportunities, pp. 112 – 119, December 2007. the purpose for which it was obtained. Such processing is [24] https://doi.org/10.1007/978-0-387-21818-2_7; Davis G.G. (2004) thus limited to what is necessary to fulfil the purposes, for Unethical Expert Witness Testimony. Springer, New York, NY. which he was appointed. If here a risk is identified, it [25] IEC 61672 – 1, Electroacoustics – Sound level meters, Part 1: Specifications. needs to be eradicated or reduced. If he want to use [26] P. Rasmussen, G. Flamme, M. Stewart, D. Meinke and J. Lankford, personal data for other reasons, he must inform the »Measuring recreational firearm noise«, Sound & Vibration, August 2009. plaintiff before doing so, and seek consent about this. [27] Jeremy R. Gaston and Tomasz R. Letowsky, »Listener perception of single shot small arms fire«, Noise Control Engineering Journal 60(3), pp. REFERENCES 236-245, June 2012. [28] M. Čudina, J. Prezelj and F. Deželak, »Noise immission from firecrackers«, Proceedings of the Eleventh International Congress of [1] B.M.Krizaj, The decision of the District Court in Ljubljana, IK 962/2004 Sound and Vibration, St Petersburg, pp. 1291-1298, July 2004. with the perpetrator's commitments, October 23th 2007. [29] F. Dezelak, J. Prezelj and M. Cudina, »Some Statistical Aspects of [2] Judgment of the High Court in Koper, Cp 381/2012, Aug 8th 2012. Firecracker Noise«, Journal of Mechanical Engineering, pp. 529-541, [3] M.L.Brecelj, Judgement of the District Court in Ljubljana, no. VI K 55(2009)9. 38124/2014, February 17th 2015. [30] R. P. Hamernik and K. D. Hsueh, »Impulse noise: Some definitions, [4] A. Bedene, Judgment of the District Court in Ljubljana II P 1669/2015, physical acoustics and other considerations«, JASA 90(1), pp. 189-196, September 24th 2018. July 1991. [5] Anton Gros, Expert witness testimony no II P 1669/2015, District [31] N. Kapoor and A. P. Singh: »Firecracker noise and its auditory Court of Ljubljana, pp. 1 – 10, August 14th 2017. implications«, Proceedings of the Tenth International Congress on Sound [6] The cross examination of the expert Gross at the District Court, II P and Vibration, Stockholm, pp. 5063-5070, 2-6 July 2003. 1669/2015; May 16th 2018. [32] O. V. Mohanan and M. Singh: »Characterisation of sound pressure [7] F. Dezelak, Influence of impulse noise transients on energy levels produced by crackers«, Applied acoustics, Vol. 58, pp. 443-449, equivalent; Ph.D. thesis, University of Ljubljana, Faculty of Mechanical December 1999. Engineering, Ljubljana 2005. [33] Military noise environment Hearing Protection – Needs, [8] K. Ceranja, B Javornik and A. Panjan, Judgment of the Higher Court in Technologies and Performance, NATO Technical report by Task Group Ljubljana, I CP 2544/2018, June 12th 2019. HFM – 147, chapter 3, November 2010. [9] R. Straus, N. Betetto and A.B. Penko, Decision of the Supreme Court [34] M. Kozinc, Court correspondence with proposals of former minister in Ljubljana, II DoR 515/2019, November 21th 2019. to the District Court in Ljubljana, II P 1669/2015, July 16th 2019, August [10] M. Pavcnik, D. J. Pensa and R. Knez, Decision of the Constitutional 23th 2019. Court in Ljubljana, II Up 96/20, March 9th 2020. [35] P. Baša, Judgment of the High Court in Koper, I Ip 45/2023, Jun 15th [11] Final decision of the Constitutional Court, II Up 96/20– April 21th 2023. 2020. [36] E. Spetič, Explanation regarding the decision on enforcement of the [12] Directive 2013/29/ EU of the European parliament and of the council District Court in Postojna, 0449 I 53/2022, July 7th 2023 and July 21th of 12 June 2013 on the harmonisation of the laws of the Member States 2023. relating to the making available on the market of pyrotechnic articles. Deželak: Ilegal use of firecrackers, Part 1 1st Author Surname et al.: Paper title 216 Ilegal use of Firecrackers and its Consequences - Case Study of Human Rights Violation at Slovenian Courts – Part II: Physical Background Ferdinand Deželak Retired researcher Abstract chemical energy is partially converted into In second part of this paper a physical background of electromagnetic energy (heat and light), and partly into firecracker explosion and its influence on hearing damage mechanical energy as a shockwave. The physical is described. This explosion took place very close to the characteristics of such explosions depend on the type of victim’s right ear. The peak of its sound pressure was firecracker and its composition, primary on the size, around 150 dBC. However, due to incorrect metrics used, weight and encapsulation of the explosive used which the expert witness estimated intuitively this level to be present the most important emission factors. Apart from emission, a considerable role in sound overpressure much lower, with four hundred times to low energy value, formation at the point of reception is played by the received by the victim’s ear. The expert completely distance and presence of different objects that influence misinterpreted several European directives concerning this reflections, diffractions, etc (Annex B). field. Based on such erroneous assumptions, he concluded Firecracker explosions belong to a group of high impulsive that no hearing damage could be possible as a result of noises which are particularly hazardous for hearing this explosion. Apart from incorrect metrics, he also impairment. The firecracker explosion produces a confused some important acoustic quantities and facts, transient sound signal which is manifested as an impulse resulting in additional cardinal errors. Due to a lack of noise with sharp rising time (more than some 100 dB/s) fundamental knowledge regarding high impulse noise and and of short duration. its propagation, he was unable to recognize the It is true, that the noise from a gunshot, especially from importance of its amplitude and spectral characteristics some larger calibers, can be even noisier. However, it is for hearing damage. He further confused reflection, important to note that shooters, bystanders and other refraction, diffraction and other factors, resulting in personnel exposed to such noise usually tend strictly to misleading expert testimony and its wrong conclusions. apply hearing protection devices. On the other hand, Despite numerous warnings and evidence, the judges when fireworks are used, such protection is almost never uncritically accepted these erroneous conclusions and applied. This is especially true for persons not using made completely unacceptable and fraudulent judgments, firecrackers and during the time when their use is strictly resulting in the conscious violation of the victim’s human prohibited. rights by various Slovenian courts. Keywordss: audiometry, corruption, criminal proceeding, 2. THE INFLUENCE OF FIRECRACKER NOISE TO HEARING expert witness testimony, firecracker noise, hearing DAMAGE damage, high impulse noise, human rights violation, litigation, peak sound pressure level. When a firecracker explodes, it produces ear-splitting sounds which may be amusing to the user, but on the other hand, this is generally recognized as a source of 1. SOME PHYSICAL CHARACTERISTICS OF FIRECRACKER danger noise pollution in the environment. This can NOISE further shift the hearing threshold or even produce deafness. Many impulse noises from firecrackers are so A firecracker usually consists of a cylinder-form tube, filled intense that a single unexpected impulse incident to with an explosive, producing a loud noise when it unprotected ears can result in severe and permanent explodes. In such explosions a definite amount of hearing loss, although the impulse is usually very short Deželak: Ilegal use of firecrackers, Part 2 217 AAAA – 2023 – IZOLA - Conference Proceedings and contains relative little energy. earmuffs or earplugs. Of course, this holds true only when Impulse noise from firecracker explosions creates several such exposure is expected in advance. For this reason the particular hazards to the human auditory system. First, additional importance of differences between expected the high peak levels associated with firecrackers may and unexpected impulses must be taken into damage the cochlea by causing rapid mechanical failure consideration. and injury. A series of rapidly occurring impulses can be High impulse noise, especially when a firecracker partially attenuated by the acoustic reflex, a reflexive explosions are involved, has a different effect on the ear contraction of the middle-ear muscles, while isolated than usual continuous noise, as the protective impulses reach the cochlea unattenuated before mechanisms of the ear are less effective for a very short activation of the acoustic reflex [20]. It is thus important noise. A 1.5 ms lasting impulse noise at 150 dB can that contraction of the middle ear muscles can occur prior damage the ear permanently to some degree even though to occurrence of the impulse noise if such a noise had its equivalent energy level wouldn't be significant. been warned prior by another loud signal. Price refers to the human reaction to such an unexpected and expected 3. DESCRIPTION OF PROCEDURES UNDERTAKEN BY THE sounds as the unwarned response and warned response EXPERT WITNESS [21, 22]. Thus, the expected occurrence of a noise impulse is much safer for the hearing organ (warned response) 3.1. Erroneous Medical Investigations than the unexpected event (unwarned response - as was the case during the plaintiff’s exposure to the firecracker First the expert witness A.G., invited the plaintiff who has explosion). This difference in human reaction may be been exposed to this firecracker explosion, to undertake caused not only by the anticipatory contraction of the some audiometric tests. However, since the beginning of middle ear muscles but also by lower general physiological these tests, the plaintiff had many problems recognizing stress within the auditory system. For example, an the audiometric signals. There was no significant sound extreme unwarned response (the startle response) is absorption in the test room, and audiometric test signals characterized by vasoconstriction and can affect the were difficult to recognize due to the high level of biochemistry of the organ of Corti. Thus, single intense background noise – see chapter III.D. A loud conversation and unexpected explosions may result in large cochlear between other clients and employees has been clearly lesions and significant hearing loss. The most common heard from adjacent areas, so the plaintiff did not have symptom of such an acoustic trauma is tinnitus, the opportunity to recognize the audio signals clearly. He manifesting as a ringing in the ears [20]. Although a certain was disturbed and distracted by non-related events and amount of hearing recovery takes place after an acoustic by people in the vicinity. After the plaintiff pointed out this trauma episode, the individual is often left with severe, issue, the expert tried to mitigate this problem by simply permanent hearing loss. So at last 7% hearing loss of the sending her nurse to the adjacent rooms in order to plaintiff was undoubtedly the result of the firecracker quieten these loud and disturbing conversations. In this explosion close to his right ear, which the expert was way the expert performed the audiometric tests under unable to explain and what was confirmed at auditory very unsatisfactory and unconventional conditions. tests. The expert did not describe the compliance with Exposure to noise levels greater than 140 dB can cause standards of the audiometric room in which he performed permanent hearing damage. Almost all firecrackers create these auditory tests. He also failed to state whether the noise with C peak level of over the 140-dB level at a maximum permitted ambient sound pressure level, Lmax, distance of a few meters. In locations where the was not exceeded in this test room. Also, he did not reverberation is present, sounds can bounce off walls and answer whether the ambient sound pressure levels were other reflecting surfaces, making noises louder and ever measured and compared with the maximum increasing the risk of hearing loss (see Annex B). permitted levels of ambient sound pressure levels; for Exposure to the sound of a firecracker can result in example Lmax according to ISO 8253-2 at a starting permanent hearing loss, meaning the exposed person will frequency of 125 Hz and a zero level of 10 dB [18]. have trouble discerning consonant sounds such as “t”, “k”, Furthermore, he did not state what kind of walls the test "s," "sh," or "p" and other high-pitched sounds. In such room in which he carried out these audiometric case it is also difficult to understand speech on social investigations had, neither their sound insulation occasions, especially when many people are speaking properties at individual frequency bands. Finally, he also simultaneously [23]. Such people may also suffer from failed to provide information on the sound absorption ringing in their ears, i.e. tinnitus. This ringing, as characteristics of this room. accompanied with hearing loss, can be permanent. The expert argued that the plaintiff would have sought It is true that exposed people can prevent hearing loss by medical help if he suffered hearing impairment due to a using appropriate protective hearing devices, such as firecracker explosion. He did not explain, however, whether such a failure in his hearing, caused by a Deželak: Ilegal use of firecrackers, Part 2 1st Author Surname et al.: Paper title 218 AAAA – 2023 – IZOLA - Conference Proceedings firecracker, would even be curable. According to another a difference is of crucial importance since the limits of a independent expert, hearing impairment of between 7 noise when assessing any risk to hearing is expressed as LC and 12%, caused by a firecracker explosion, cannot be peak, that is using the peak value and weighted with filter C rectified. [13]. Obviously, the expert does not distinguish between these basic acoustic descriptors, denoted by LAI max and LC 3.2. Incorrect Procedures and Misuse of Plaintiff/s peak. The difference between these two indicators is enormous and in the vast majority of cases, including this Sensitive Personal Data one, affects the result decisively. During the cross The informations used by the expert were totally examination the expert only insisted, that these are all defective in order to draw any trustworthy conclusions. “audiologic decibels”, without any further explanation, Instead of professional access, he operated with quite convincing the judge which disabled any further questions irrelevant data, like the description of the plaintiff’s oral after that. cavity colour, his tongue frenulum, his throat and the The difference between the LC, peak and LAI, max indicators is throat mucous [5]. Not only are such informations about 26 dB in the actual case of the firecracker explosion. unnecessary for this kind of testimony from an expert This means that a maximum noise level of 124.08 dB(A, witness but, by doing so, he also revealed sensitive Imp) corresponds to a much higher level of LC, peak, which personal data concerning the plaintiff’s health exceeded 150 dBC at the location of the plaintiff’s most unnecessary [14]. From the existing medical exposed ear. documentation he additionally fully revealed quantitative Here the expert made a cardinal mistake which was data about the plaintiff/s blood sugar, blood pressure, probably due to a lack of understanding the most triglycerides and cholesterol levels. On the other hand, he important issues and fundamentals of acoustics. completely omitted to correlate these values with Misunderstanding these decibel units, he tried to eventual plaintiff’s hearing damage. These data were not calculate the sound pressure level at a distance of 5m for required and were totally unnecessary as part of an the F2 category firework and obtained 124.08 dB [5]. He expert’s testimony, neither did the expert use them in any did not explain which kind of decibels he had in mind. part of his testimony. Based on this incorrect noise descriptor, he simply concluded that being 5 meters from such a firecracker 3.3. Wrong Metrics Used explosion, the sound pressure level due to this firecracker explosion did not reach the limit value (130 dB) at which According to [12], the safety distance for fireworks for hearing loss could occur. He did not transform these category F2, is set to be at least 8m away. For category F3, metrics to the C-weighted level. Furthermore, he did not fireworks must be ignited at least 15 m away and the explain (and obviously even did not understand) which maximum noise level at these distances must not exceed time dynamic he used for the limit value. In his confusion 120 dB (A, imp), or an equivalent noise level determined he simply denied any possibility of hearing loss or tinnitus, by another appropriate measuring method, at the safety since, in his testimony, the plaintiff had not been exposed distance. to a impulse noise of 130 dB or more. Of course, his It is obvious that there is not an arbitrary 120 dB limit conclusion was absolutely wrong, based on his incorrect (without any dynamics and weighting specification) as the calculation and the totally wrong use of the descriptors expert used it, resulting in his quite wrong results and and their units. As already proved the L conclusions [5]. First, It is ridiculous what the expert did C peak level exceeded 150 dBC in this case. with this value! He used a point source model with According to Slovenian and European regulations, the reference 120 dB (A, imp) at a distance of 8m and cornerstone of hearing protection criteria against high calculated what the sound pressure level at 5m (the impulse sounds is a peak permitted level, L distance of the firecracker explosion to the plaintiff’s ear, C peak. This stipulates that there is a risk of hearing loss above 135 dBC according to the expert’s testimony) should be. He with respect to the reference pressure of 20 μPa. This obtained 124.08 dB (without any specification?!). corresponds to a sound pressure level of ppeak = 112 Pa. Secondly, neglecting the true limit of 120 dB (A, imp), he This means that at the time of the firecracker explosion, simply ignored dynamics and frequency weighting and the plaintiff was exposed to at least 32x (over thirty times) compared this value with hearing damage limit directly. greater sound energy level than it is permitted which, of Then he drew the conclusion that this is even less than course, resulted in acute hearing impairment. For impulse 130 dB (neglecting again any specification?!), so according dynamics, the time constant, or the rise time is 35 ms to such reasoning, the plaintiff was not exposed to any risk during its increase [25]. The time constant is defined by to his hearing at all [5]. It is quite clear that decibels, the IEC standards. It also denotes the time needed to measured with the dynamics impulse and weighted with reduce the amplitude of an exponentially decreasing filter A, are not identical with peak level decibels weighted signal by a factor of 1 / e = 0.3679... (see Annex A). with filter C or any other specification (see Annex A). Such Deželak: Ilegal use of firecrackers, Part 2 1st Author Surname et al.: Paper title 219 AAAA – 2023 – IZOLA - Conference Proceedings The expert witness, A. G., was not familiar with these formation of a tinnitus [5]. These are again speculative decisive facts and obviously confused the LPeak with an and false statements, without any evidence, and also alternative descriptor LImax, which led to enormous errors. contrary to the theory of high-energy impulse noise. The peak detector LPeak on a sound level meter must have The plaintiff here clearly explained and emphasized that a response of less than 100 μs (100 microseconds). LPeak is he was actually in the artillery, but in the computing units usually measured in combination with C or unweighted far away from artillery fire operations. Additionally, (flat) network. This is one of the most important reasons during his service, only artillery firing simulations were why the interpretations of the expert witness, A.G., are performed, so he was never exposed to real artillery completely wrong. gunfire himself. Even if he were, his hearing damage Even using a rough comparison of the time constant for would be different. Namely, the sound impulse bursts of the impulse dynamics and the peak response time, an the fireworks is quite different from that produced by approximate ratio of 350: 1 is evident, which corresponds artillery fire with a frequency spectrum that is markedly to the logarithmic ratio of about 25 dB. Additionally, the shifted towards higher frequencies, as proven in annex C. difference between the C weighted and A weighted filters Also, a comparison of these physical facts and the also makes a little contribution to this difference. It is well plaintiff's hearing loss characteristics shows that his known that C weighted levels are generally higher than A hearing impairment cannot be the result of artillery fire weighted ones, with the exception in one part of the high noise as the plaintiff had already explained that he had frequency audible region. However, since the spectrum of never been exposed to it. Also, medical evidence proved the firecracker explosions is shifted towards higher that the plaintiff's damage to hearing had increased by frequencies, this contribution is less than the contribution more than 7% between the two audiometric tests, in less due to different time constants. than 2 years apart, soon after the firecracker explosion; It is quite clear that the expert witness, A.G., confused on the other hand, this firecracker accident occurred more peak level Lpeak with the maximum level LI, max. Peak is, by than 30 years after the plaintiff had completed his military the IEC definition, the greatest absolute instantaneous service in the artillery computing units. Furthermore, the sound pressure during a given time interval. On the other default assumption of the expert, A.G., that all soldiers hand, LAI,max is the maximum level with A-weighted and other personnel in artillery are exposed to artillery frequency response and Impulse time constant (see noise is ridiculous; this is similar to saying that all workers Annex A). employed in court are judges or all workers employed in Additionally the expert here underestimated the risk health care are doctors. Using such a set of incomplete impairment as well by using the firecracker of the F2 and mainly erroneous information, the expert witness category, although most probably a type of firecracker of tried to speculate throughout his testimony. By ignoring the F3 category was used in this case, whose peak levels most of such important data and lacking even the most greatly exceed 155 dBC at a distance of 5m. fundamental theoretical knowledge he was unable to prove his claims of course. 3.4. Confusion about Sound Wave Phenomena The expert witness considered the unevenness of the 3.6. Confusion and Misunderstanding of Occupational walls and the corners of the surrounding houses where, Noise Exposure according to his testimony, the air waves were refracted The expert witness further assumed that the plaintiff was [5]. There were obviously no air waves in question, but occasionally exposed to high levels of noise at his rather sound waves. Even worse, he completely confused workplace. He further speculated that the protective the effects of reflection and refraction with diffraction and hearing devices were not sufficient as the noise was also scattering. The expert witness obviously does not transmitted through the cranial path to the inner ear [5]. distinguish between these terms [5, 6]. However he did not provide any evidence about the propagation of the sound energy transferred to the inner ear through the bones of the skull against airborne 3.5. Confusion about Low and High Frequency Content of transmission. Although he tried to explain hearing loss in an Explosion this way, he was unable to give any estimate of how much According to his biased position, the expert was looking of the sound energy was transferred to the inner ear for some unrealistic facts in trying to explain the plaintiff’s through the bones of the skull against airborne hearing loss. He namely declared that the plaintiff may transmission. These expert statements were thus again have suffered hearing impairment during his military baseless. The use of earplugs and earmuffs (especially service in the artillery more than 35 years earlier. Without when combined) successfully protects or reduces the any evidence he guessed that he was certainly repeatedly exposure of auditory organs to noise levels by up to 40 dB exposed to strong impulse noise and hearing loss which in the frequency region of interest. Only when even higher consequently would, in his testimony, result in the reductions are required it is necessary to reduce the Deželak: Ilegal use of firecrackers, Part 2 1st Author Surname et al.: Paper title 220 AAAA – 2023 – IZOLA - Conference Proceedings transmission to the inner ear by the cranial path as well. other hand, the right ear is located in the shadow of the Neither the plaintiff nor the vast majority of workers have head, which protects it, particularly at high frequencies. need to take any of such precautions so far. The plaintiff The ear that is tucked into the shoulder thus sustains suggested the expert to describe one practical case lower dB levels than the ear that is exposed to the muzzle where, according to his experience, such or a similar of the gun, so the left ear sustains higher levels of damage problem arose in his practice, but he was unable to give than the right ear in right handed shooters. When any answer at all again. shooting right-handed (as in the case of the plaintiff) the In the case of any excessive noise, or even on a suspicion shooter’s left ear is facing forward and thus receives more of it, the plaintiff always used personal protective of the muzzle blast energy than the right ear which is more equipment to protect his hearing; if necessary even backward facing. On the other hand, the plaintiff, as most additional protection, by which the noise emission levels shooters, used correctly fitting personal hearing devices in the ears were reduced by up to 40 dB. This was also his for ear protection to both ears. duty, as he provided a professional service in this field, This fact is indisputably confirmed by a number of such as noise assessment and its control. So it is quite clear investigations and statistics regarding hearing damage of that different occupational and recreational types of noise shooters, especially hunters who rarely use personal to which the plaintiff was subjected to through his duties protective equipment. Such asymmetry in noise exposure presented no risk of damage to his hearing. The emission levels of sound energy released during occasional bursts applies practically to all shooters (except to left-handed and similar noisy events which he received during his ones, who mostly use customized firearms especially workshift, were consequently approximately ten adapted for them). A similar phenomenon occurs with thousand times lower than that which the plaintiff's inner many professional musicians as well, especially violinists ear received during the firecracker explosion. On the and string players. It was consequently proved that the other hand, the plaintiff found himself in a quiet different left ear predominantly experienced more hearing loss situation on April 10, 2012, as he was been completely surprised by an unexpected firecracker explosion. So he during sport shooting activities than the right. For this did not have any chance to take the appropriate reason hearing loss amongst rifle shooters tend to be preventive measures or to use any other form of asymmetrical, since the left ear is closer to the barrel and protection for his health. tends to be more impaired as it is closer to the explosion The expert witness further insisted that during whereas the right ear is additionally protected by the head recreational shooting and, at some workplaces, strong [26]. vibrations are transferred through the bones to the inner In table I there is an example presenting average C- ear. However, he was again unable to clarify what he meant by “strong vibrations”, even not in which units weighted peak pressure levels in dB when using three kind these vibrations are expressed and what their amplitude of weapons at both shooters ears [27]. should be. TABLE I C-weighted peak pressure levels in dB at both shooters ears using three 3.7. Deficient Expert Witness Testimony Regarding kind of weapons Asymmetric Exposure during Sport Shooting Weapon CZ-75 M4 M14 origin Czech United States United States The expert witness further speculated about information Republic of the plaintiff’s involvement in some sport shooting Cartridge 9x19 mm 5.56x45mm 7.62x51mm activities without any professional background [5]. Muzzle 360 m/s 880 m/s 850 m/s For this reason a few words about sport shooting activities velocity Barrel length 4.7// 14.5// 22// should be mentioned first. Here, the expert witness fully Left ear 156 dBC 156 dBC 154 dBC misinterpreted the information about the plaintiff’s Right ear 151 dBC 150 dBC 149 dBC involvement in sports shooting. The applicant was actually engaged in some occasional sports shooting activities. It is clearly evident, that the greatest level was measured However, there are some facts in relation to shooting at the left ear. The shooters were namely right handed itself that must be elucidated. During sport shooting activities, there is namely a pronounced asymmetry in the and thus the tucking of the head to the right shoulder is exposure and the impairment risk of both ears [26]. expected to provide a significant attenuation to the right During shooting, the right part of a shooter head rests ear. slightly on the right shoulder and the left ear is directed A quite opposite asymmetry was postulated due to the against the barrel that is on the side of the explosion (Fig. explosion of the firecracker being close to the plaintiff’s 1). Consequently, the left ear is much more exposed to the right ear during the absence of personal protective resulting sound impulse burst than the right ear. On the Deželak: Ilegal use of firecrackers, Part 2 1st Author Surname et al.: Paper title 221 AAAA – 2023 – IZOLA - Conference Proceedings devices, exposing his right ear to the noise of the Court were under his pressure as well. In this way the case firecracker. acquired a completely new political and corruptive Even the expert A. G. himself, confirmed that the dimensions. plaintiff’s right ear is significantly more damaged than the Here, a very important role played a wrong expert left one. This assertion additionally confirms the fact that testimony as appointed by judge A.B.. This expert witness the plaintiff was unable to obtain such hearing damage as heavily strayed from the truth in his testimony. The a result of some sport shooting activities, but as a result of expert’s testimony fell deeply below the reasonable a powerful and unexpected firecracker explosion which standard for his profession. occurred in the immediate vicinity of his unprotected right In the existing expert testimony, there are a number of ear. significant errors, many of them of a fundamental nature, which have a decisive influence on the final results of this 4. CONCLUSION expert testimony and its conclusions. The expert performed audiometric tests under extremely It is undisputed, and even the attacker himself did not unfavourable conditions; he paid no attention to the deny the fact, that he deliberately threw an explosive fundamentals of acoustics and audiology, nor to the device on the plaintiff, causing permanent health human perception and reaction to sound. He confused the damages to him. It was thus proved beyond a reasonable acoustic characteristics of firecracker explosions with doubt that the perpetrator committed the crimes artillery and sports firearms. The expert did not take into charged, which consequently caused damage to the account the basic acoustic characteristics of such different plaintiff. Additionally, this has been confirmed by the police investigations as well. In this way, the courts groups of bursts. Furthermore, the expert witness intentionally ignored all important evidences and proofs. disregarded many physical laws and facts, such as Almost all criminal proceedings were corrupted and differences in the spectral characteristics of small completely biased at the District court in Ljubljana against (firecrackers) and large explosions (artillery shells). He the attacker. The former Minister of Justice with his neglected the sound exposure asymmetry characteristics political influence further managed to convince the Higher of the left and right ear during sport shooting and the Court in favor of the attacker. Despite a proven attack with efficiency of personal protective equipment used during explosive devices the attacker has been fully acquitted in a workshift, etc. criminal trial. The most fatal error he made, however, was the confusion Similar misconduct was processed during the lawsuit. The between different metrics he employed in the evaluation lawsuit proceedings were conducted in a biased manner, of high-impulse noise. He simply confused the descriptor starting with a District Judge F.K.. He was known for some LAI, max and LC, peak, which differs by approx. 26 dB for of his scandals, which are completely incompatible with firecracker explosions. This additional confusion in metrics his judicial function. Since he violated his judge's resulted in a fatal error of the peak level by 26 dBC. This obligations of impartial conduct more times, he was fired. difference was even a little higher, as the firecracker Before that however, this judge delayed the case for applied in his calculation was of a third, rather than of a almost five years in favor of the perpetrator. Despite this second category (Megatron or Extreme of type 3 instead punishment the District Court still persistently conceals of Pirate of type 2, with which the expert based his from the plaintiff and other publicity the details about this calculations). In this way, the expert’s conclusion judge and his corruption. regarding the exposure to 124 dB peak level instead of a The case was then handed over to a new judge, A.B., who much higher value of over 150 dB was totally wrong of was also immediately pressured by the former Minister of course. It is ridiculous that all judges involved still have Justice. She did not condemn the planned attack on the faith in such erroneous data and conclusions. plaintiff, with a dangerous explosive device. Even more, Almost all expert witness conclusions were consequently she even approved this attack on the plaintiff's body, completely wrong due to the substantial errors in his despite quite opposite international and slovenian law calculations, his incorrect application and practice. A former Minister of Justice M.K. namely intervened many times in the attacker's favor. In this way, misunderstanding of the metrics used. the case completely deviated from the established Apart of criminal and lawsuit legislation many Directives practice of law and norms, just in order to protect the and standards were violated at different slovenian courts. attacker interests for any price. All the plaintiff's appeals Such infringement present a serious violation of human were in vain. Indirectly, the Supreme and Constitutional rights and dignity of the plaintiff. Deželak: Ilegal use of firecrackers, Part 2 1st Author Surname et al.: Paper title 222 AAAA – 2023 – IZOLA - Conference Proceedings It is clear that judges must be accountable to legal and and Lpeak. Contrary to the expert A.G. opinion, these are quite ethical standards in making an admissibility decision on different quantities. expert testimony and other lawsuit procedures. However, they were not. Noise levels can namely be measured or expressed with different dynamics, each with a specific time constant. In practice, three such dynamics are mostly used today, ANNEX A– THE IMPORTANCE OF DIFFERENT TIME which are most often built into modern noise meters: WEIGHTINGS Fast, Slow and Impulse [25]. When measuring the noise level, we actually detect A maximum sound level is the highest exponential -time- changes in the sound pressure caused by the sound average sound level, in decibels, that occurs during a disturbance under investigation (in our case, a firecracker stated period. For a given nonsteady sound pressure waveform, the maximum sound level depends on explosion). Such sound pressure levels change very exponential-time-weighting, which is used (fast, slow or quickly, so that their actual reading with a real-time impulse). measuring instrument would be very difficult. As a result, The maximum A-weighted noise level measured with the measuring instruments are fitted with certain dampers impulse dynamics presents the maximum effective (RMS) that dampen their response to such rapid time changes value, evaluated by the A filter weighting and measured and allow for a smoother reading. Such a process is called by the dynamics with a time constant of 35 ms. time weighing. The RMS level describes the effective value of the square The IEC 61672-1 standard [25] deals in more details with of the sound pressure, denoted as pRMS, and is expressed two different time weightings, Fast and Slow. Both are in pascals (Pa). The RMS value is thus calculated by characterized by a certain damping, or a slow reaction of squaring the value of the time-dependent sound pressure p(t), its averaging (integration) over a certain time interval reading the screen of the noise meter. Dynamics Fast of t course reacts faster, its damping is determined by the 2-t1, and finally its root square is taken, which is mathematically written as: time constant τ = 125 ms; the dynamics slow is determined by an eight times larger time constant τ = 1 s. 1 𝑡 If, on the other hand, the sound stops suddenly, the signal 𝑝 2 𝑅𝑀𝑆 = √ ∫ 𝑝(𝑡)2𝑑𝑡 (1) 𝑡2−𝑡1 𝑡1 on the display starts to decrease rapidly at 34.7 dB / s for In (1) t1 and t2 are the time moments of the beginning dynamic fast. In the case of slow dynamics, this decrease and end of the evaluating interval of this average. is eight times slower, or 4.3 dB / s [7]. It is therefore obtained by an integration of the squared Figure 2 shows the response of the measuring instrument value of the sound pressure, its time averaging and finally to a rectangular sound signal at both described dynamics. taking the square root of this value. Simple averaging would not give a representative value (for example, in the case of a sine wave, this value would be practically zero), while the RMS value is proportional to the energy of the sound signal, which is important to describe its effect on hearing impairment. On the other hand, the peak level is, according to the IEC definition, the highest absolute level of the current sound pressure during a given time interval. Figure 1 shows the Fig. 2 Fast and slow dynamics properties: difference between the maximum Lmax and the peak 1-real time sound level, 2-displayed Fast level, 3-displayed Slow Lpeak level. It is clear that, contrary to the expert witness level opinion, these quantities are not identical. For measurements of short impulse noise events (as in the case of a firecracker explosion), measuring instruments are usually equipped with a third type of dynamic - Impulse. However, in contrast to the described dynamics Fast and Slow this dynamic is asymmetric. Thus, the pulse dynamics adapts very quickly to the increase of the sound signal (τ = 35 ms), while its decrease after the termination of the sound signal is much slower, namely only 2.9 dB/s Fig. 1 Graphical representation of the difference between Lmax Deželak: Ilegal use of firecrackers, Part 2 1st Author Surname et al.: Paper title 223 AAAA – 2023 – IZOLA - Conference Proceedings with a tolerance of ± 0.8 to ± 1.3 dB/s, depending on the time constant τ yields a running time average of the measuring class system. Figure 3 shows the response time frequency-weighted, squared sound presssure. The to a sudden sound pulse at the slow, fast, impulse, and peak running time average indicates that an exponential-time dynamics. average sound level displayed by a conventional sound Indicators with time impulse evaluation are considered less level meter is continuously updated as a function of appropriate today and are not recommended for assessing observation time. the risk of hearing impairment [25]. In any case, its use In the above equation, the numerator of the logarithmic requires appropriate knowledge and experience. argument is the exponentially weighted and frequency- evaluated rms value of the sound pressure at time t. The standard design goals for fast and slow time constants are 0.125 second and 1.000 second, respectively. Time constant τ is equal to the time required for a quantity that varies exponentially with time to increase by the factor [1- (1/e)] or to decrease by the factor 1/e, where e is the base of the natural logarithm (e=2.71828). The output of the sound pulse signal from the detector is thus most simply described by two functions, depending on whether it is increasing or decreasing. The increasing Fig. 3 Response time to a sudden sound pulse at different signal is illustrated by an increasing function with zero dynamics: slow, fast, pulse and peak: 1-Peak, 2-Impulse, 3-Fast, input according to the equation 4-Slow t - (3) L(t) 10 = log e - (1  ) Mathematically, time evaluation is usually expressed by an exponential function of time, or the corresponding This is an increasing transcendent function with a positive time constant. This represents the weighting of the square first and a negative second derivative, with no input signal of the current sound pressure. The time-weighted sound at the initial moment. level, however, is defined in accordance with the IEC The decaying signal after the termination of the pulse is 61672 [25] standard as the twenty times of a common illustrated by a uniformly decreasing function with zero logarithm of the ratio between the square root of the output, [7]: t square of the sound pressure and the reference sound L(t) 10 = log ( e- ) (4) pressure. An A-weighted sound pressure level in decibels of an A- which actually represents a linear function with a negative frequency weighted sound pressure signal and for first derivative. exponential-time-weighting e-(t-ξ)/τ is, at any observation Figure 4 also shows a practical example of the responses of time t, determined according to different noise dynamics to a pulse signal, which is relatively common in practice. 𝑡−𝜉 2 1 𝑡 − ) ∫ 𝑝 2(𝜉)𝑒 𝜏 𝐿 𝜏 𝐴 −∞ 𝐴𝜏(𝑡) = 20 log √( 𝑑𝜉 (2) 𝑝0 where p2A(ξ) is the square of the instantaneous A-frequency- weighted sound pressure in pascals τ – a time constant, for a selected time weightings F, S or I ξ – a dummy variable of integration in time interval between -∞ and t p Fig. 4 A practical example of the responses of different time A(ξ) – the instantaneous A-frequency-weighted sound weightings to a impulse signal: 1- Impulse, 2-Fast and 3- Slow pressure in pascals p0 – the reference sound pressure of 20 μPa. Running integration here occurs from some very distant ANNEX B. TYPES OF IMPULSE NOISE DUE TO FIRECRACKER start time (theoretically at -∞) to the present time t [7]. EXPLOSIONS Dividing the result of the integration in above equation by Deželak: Ilegal use of firecrackers, Part 2 1st Author Surname et al.: Paper title 224 AAAA – 2023 – IZOLA - Conference Proceedings A firecracker explosion outdoors, in an open space  t − t  o without any reflective obstacles, produces a non- p ( t) = p for t  t  r peak o 1 t   (6a)  − 1 t to  reverberant A-type impulse noise with a single spike-form overpressure which can be approximated as a Friedlander and t − t impulse as shown in Fig. 5 a and 6. On the other hand, a 1  − t −  1 t t 2 1 p (t)= p 1 − − e t for t  t firecracker explosion indoors or in a semi-open space p peak 1   (6b) t  − 2 1 t  results in a reverberating effect and are usually described Here p peak is the overpressure amplitude and t 2 is the time as a B- or C-type impulse, Fig. 5 b and c [28]. taken for the overpressure to fall to zero value for the first time. In mathematical form the A-type of impulse is often described by the Friedlander pulse, starting with rapid rise from ambient pressure to a peak level, after which its more slowly decay follows back to the ambient pressure [29]. Usually, such kind of impulse appears as a result of explosion in an open environment with no reflecting surfaces. Fig. 5 Sound pressure levels of different types of impulse noise The peak value of sound overpressure is the maximum in time domain; a) A-duration, b) B-duration and c) C-duration absolute value of the instantaneous sound pressure in Pa. The rise time is the time interval between the start of an These types are characterized through rising time, impulse impulse and the time when the peak value is attained. Due duration and peak overpressure as the main parameters to practical reasons, this is usually taken as the time describing impulse noise [29]. These parameters depend required for the sound pressure to rise from 10% to 90% on the source, i.e., firecracker used. of its maximum absolute value [29]. Duration is the time The duration of A-type impulse (also called A-duration) is interval between the start of an impulse and the time the time required for sound pressure to reach its unweighted peak value and then to fall for the first time to when an impulse decays to a zero value. zero value, ( t All of these parameters (peak level, rise time and 2-t o), Fig. 6. In the case of ideal waves, it is equal to the positive phase duration used in blast physics duration) depend on the type of firecracker and its [29]. With B-duration, the total time required for the peak characteristics, such as its mass, charge, the type of of a pulse level exceeding the criteria -20 dB, is depicted explosive used, encapsulation, its geometry and design in Fig. 5b. C-duration is determined as the time required [29]. for the envelope of the unweighted peak sound pressure to Typical firecracker explosions outdoors, at a distance of decay by 10 dB (Fig. 5c). The corresponding level is some meters, produces a peak sound pressure level highly therefore about two-thirds of the peak pressure value. exceeding levels as adopted for hearing protection. A The A-type impulse can be described as a combination of typical firecracker exploding outdoors produces high linear function for the rising part and exponential function sound pressure levels [31, 32], at distances of around 5m, during its decay [30], (fig. 6): peak overpressure levels is between 145 and 160 dBC. The spectral distribution of energy in such an impulse is controlled by the A-duration and the rise time. Where there are reflective surfaces, such as the ground or walls, additional pulse components can appear (see Fig. 5 b and c). Reflecting surfaces produce secondary reflected waves. These reflections can interefere with the original pulse, producing a complex temporal pattern. The time interval between the primary and secondary waves depends on the relative distances between the reflecting surfaces and the receiver. The peak and spectrum of the reflected components depend on the impedance characteristics of the reflecting surface. Fig. 6 The Friedlander wave as a firecracker explosion in time domain ANNEX C – FREQUENCY CHARACTERISTICS OF DIFFERENT p( t) = p ( t) + p ( t) (5) EXPLOSIVE SOURCES r p where When launching artillery missiles, one has to deal with Deželak: Ilegal use of firecrackers, Part 2 1st Author Surname et al.: Paper title 225 AAAA – 2023 – IZOLA - Conference Proceedings substantially larger masses of explosives, exceeding Looking for the extreme value of this function, yields 2 4 thousands of times the mass of firecracker explosives. For d | 2 P   F ( ) | t p - t (11) = p 0 = 2 2 2 this reason, the duration of artillery explosions is d (1+ t p ) considerably longer than that of firecrackers, resulting in from which it follows lower frequency components. 1  2 =  f = (12) F In figure 7b and 8b the time dependence of sound t p overpressure is shown which is an approximation of sound or f F = 1/2π t p (13) impulses resulting from firecracker explosions and heavy It is evident that the spectrum reaches the maximum weapon operations. amplitude at the frequency f Here the rising time is very short, so only the exponential F = 1/2π t p, and the maximum function during the decay (eq. 6b) of the Friedlander value of the amplitude spectrum is  P()max = t p/2. impulse is of practical importance. In the time domain, the Therefore, there is a significant difference between the sound pressure of the Friedlander pulse p duration of these two different explosions. For example, a F(t) can be written in the form firecracker produces an impulse with an A duration of about 1ms and a peak in the spectrum at about 2000 Hz. t − t1  t − t −  1 t − 2 t1 p (t)= p 1 − (7) e At launching an artillery shell, it has a duration of about F peak    t − t  2 1  ten times longer so the maximum spectrum level is shifted where p peak is the maximum peak sound pressure. to lower frequencies by one decade, which is around 200 Hz (figure 7c and 8c) The time domain can be transformed into a frequency domain by using the Fourier transform P F( ω) [7, 28]:  i t2p  P (  = ) p (t ) F F e-i t = dt  (8) (1 + i 2  t ) p - (a) (b) (c) where ω = 2π f is the circular frequency. Fig. 7 Firecracker explosion characteristics a), time diagram of Next, we find the effective value of pF(rms). sound pressure b), shape of its spectrum shifted to high 2 frequencies [33] c) T 1 1 -2T/ t p t p t p 2 P T P T P (rms) =  (t F p ) dt = ( e ( - + - + ) ) F 0 2t 2 4 4 p P T P T (9) whereby the TP is selected as a point on the time axis in which the pressure p F( t) is asymptotically approximated to 1% of the baseline value, which is approximately TP = (a) (b) (c) t r+6 t p. Here TP denotes the time required for such an Fig. 8 Howitzer firing characteristics a), time diagram of sound approximation, where the time of rising is t r and t p is the pressure b), the spectrum is shifted to low frequencies [33] c) time of the decreasing Friedlander pulse to zero. Thus the sounds produced by large-caliber weapons have In this way the frequency spectrum of the pulse or its acoustic energy predominantly concentrated in the low energy distribution is obtained. The amplitude spectrum frequency range (below 400 Hz), while the spectral is given by the following equation [30] content of sounds produced by firecrackers extends 2  t | p P  (10) higher, around 1500 – 2500 Hz. F ( ) = | (1+ 2 2  t p ) Deželak: Ilegal use of firecrackers, Part 2 1st Author Surname et al.: Paper title 226 LOW FREQUENCEY NOISE MEASUREMENT IN THE PASSENGER CABIN Domen Bartolj; Samo Beguš University of Ljubljana, Faculty of Electrical Engineering, Tržaška cesta 25, SI-1000 Ljubljana, Slovenia Abstract: In recent years, the negative impact of sounds of infrasound and ultrasound frequencies on the health of people has been observed and documented. There are more and more sources of infrasound (e.g. traffic, wind farms) or ultrasound (e.g. electronic mole repellers, remote controls, sensors) in everyday life, but at the same time there are few recommendations and safety standards governing this area. Therefore, research on the impact of ultrasound and infrasound on humans is very important and necessary. A case study of the low frequency noise measurement in the passenger cabin is presented. A sound level meter with a separate microphone is used to assess the noise level in the passenger cabin when driving with closed or open window. Post processing is carried out later with different commercial and freely available software. The results show sound pressure levels of more than 100 dB and 110 dB in a car cabin with closed and open window respectively. Keywords: Infrasound, low frequency noise, sound pressure level measurement, passenger cabin, Room EQ Wizard, LabVIEW 1. IMPACT OF INFRASOUND ON HUMAN HEARING infrasound and low-frequency noise can lead to a sensation of pressure in the ear. Extensive research has People perceive the spectrum of frequencies with their shown that this effect of infrasound begins to occur at auditory system in the frequency range from 20 Hz to infrasound levels between 127 dB to 133 dB and does not 20,000 Hz. This range varies among individuals and can necessarily increase proportionally with rising sound change with age, hearing disorders, and exposure to pressure[1], [2]. noise. Infrasound is generally not detectable by the naked The impact of low-frequency noise on communication can ear but it can be perceived by increasing the sound be quite pronounced. At high intensity levels of low- pressure level (SPL) [1]–[5]. frequency noise, effects have been reported such as voice Exposure to low-frequency noise can lead to irritation and modulation, feelings of muffled sound, and chest wall even damage to human hearing system. In this regard, vibrations [2]. numerous tests have been conducted. These tests include There is no clear evidence from experimental studies to the threshold of audibility, which has been tested by many support claims that infrasound has a negative impact on researchers, including frequencies below 20 Hz [1], [2]. human performance, induces a "drunken" state in Another test, known as the Temporary Threshold Shift individuals, or directly triggers nystagmus. The effects (TTS), examines situations where prolonged exposure to observed at intensity levels ranging from 105 dB to intense low-frequency noise can temporarily cause 120 dB, if they can even be validated, are likely to have reduced audibility or even permanent deafness [2], [6]. been exaggerated [7]. However, it is not precisely known Ear pain and injuries are related to the mechanical at what level infrasound becomes dangerous to humans movement of the middle ear beyond its normal and how it affects their physical abilities. operational limits. The pain threshold is not well defined Infrasound sources could be of natural or artificial origin. and is different for each person. Furthermore, exposure to Natural sources of infrasound include earthquakes, Bartolj et al.: LOW FREQUENCEY NOISE MEASUREMENT IN THE PASSENGER CABIN 227 AAAA – 2023 – IZOLA - Conference Proceedings tsunamis, storms, and animals (mostly water animals). For noise measurement and data recording the Brüel & Artificial sources of infrasound include industry, traffic, Kjær Type 2270 Sound Level Meter with Type 4189 various electrical devices, wind turbines, and human prepolarized condenser microphone was used. activities in construction works, among other things [1], For a reference SPL calibration the Brüel & Kjær Type 4231 [4]. Class 1 Sound Level Calibrator was used. Artificial noises have a more significant impact on people Data processing was performed with LabView and a freely than natural ones, as people react to them more strongly, available software Room EQ Wizard (REW). and in everyday life, we are exposed to them more. An example are wind turbines, as they indicate that the 1.3. Procedure and setup prevalence of low-frequency noise is a particular concern for communities living near these power plants. Today's The process of conducting the measurement in the Clio low-frequency noise, primarily originating from Storia car involved installing a microphone, which was machinery, can lead to perceiving it as a constant placed on the headrest of the passenger seat and background noise but at night, when ambient noise connected to a sound level meter placed on the passenger decreases, low frequency noise dominates. Especially seat. during the night, the low frequency noise could be perceived by humans [1]. Because we are constantly exposed to low-frequency noise and noise in general, environmental noise is measured. The measurements are conducted in accordance with the ISO 1996-2 standard. Additionally, legal regulations define the limits of permissible noise levels at different times of the day and in various environmental zones [8]. The largest source of noise pollution nowadays is traffic [9]. Noise, vibration, and harshness (NVH) is used to evaluate the noise and vibration characteristics in vehicles. The measurements conducted for noise and vibration are objective, whereas the harshness measurements are subjective, as they are evaluated by a group of evaluators or provided through analytical tools. Fig.1. Measurement equipment set-up in the car. They deal with the noise and vibrations experienced by passengers inside the cabin. External measurements focus The microphone was set up for measurement in the on the noise emitted by the vehicle. Passengers are diffuse field environment. With this selection, the sound subjected to the noise, which depends on the level meter also automatically sets correction filter for environmental conditions (car type, driving speed, car each microphone, optimizing its frequency response [13], resonances, open or closed windows, etc.) [10]–[12]. see Table 1. In the present article a measurement and analysis of a car cabin noise is performed to assess the conditions when driving. The SPL and spectrum of the cabin noise was measured. Table 1. Free-field frequency range for microphone type 1.2. Equipment used 4189 [13]. The measurement were carried out in the Renault Clio Before starting the measurement, we calibrated the Storia 1.2 car. sound level meter using a sound calibrator. Location logging and driving speed measurements were recorded by a telephone GPS receiver [14], [15]. Bartolj et al.: LOW FREQUENCEY NOISE MEASUREMENT IN THE PASSENGER CABIN 228 AAAA – 2023 – IZOLA - Conference Proceedings After the calibration we drove the car along the streets of Using LabView data processing and data flow is presented Ljubljana, including the highway. Our journey commenced In Fig 3. LabView is able to read directly from a wav file. in front of the Faculty of Electrical Engineering in The signal read from the file is frequency weighted, band Ljubljana, starting on Barjanska Road, then transitioning pass filtered (if we want to limit the frequency range of onto the highway until the Brdo exit, followed by driving interest), time weighted and plotted. Car speed data is along the road towards Brdo, see Fig 2. also read from the file and visualized in a plot. During the measurements, the recordings were saved as WAV file for later analysis. 2. MEASUREMENT RESULTS Before starting the measurements, it was necessary to calibrate the sound level meter. We accomplished this by using the B&K sound calibrator type 4231. We performed calibrations at SPL levels of 94 dB and 114 dB. Spectrum at 114 dB is shown in Fig. 4. The reference SPL is also visible in Fig. 5 and Fig. 6 in the time interval from 1:00 minute to 1:20 minute. Fig.2. Presentation of the driving path 1.4. Data analysis Post processing and visualization of the data was performed using a commercial program LabView [16] and with a freely available program REW [17]. For the playback, we employed the freely available program Fig.4. Reference 114dB frequency spectrum in REW. Audacity [18] and utilized the VB-Audio Virtual Cable [19] to avoid linear frequency distortion and reduction of During the measurement, the sound level meter stopped signal to noise ratio to feed the signal from recorded wav logging the data for approximately a minute and therefore file to the program REW. there is a 1 minute long loss of data, which can be seen in Program REW can utilize a SPL logger to visualize the SPL the Fig. 5 and Fig. 6. The average SPL at lower speeds was change in time, display a signal spectrum and use approximately 95 dB. When the car stopped, this level frequency weightings. dropped to 80 dB. Peaks on the graph are due to the To calibrate SPL levels in LabView and REW the reference nearby traffic. SPL levels of 94 dB and 114 dB recorded at the beginning However, on the highway at higher speeds, the SPL of the measurement session were used. without frequency weighting is more than 100 dB. The highest SPL can be observed with open windows, when the SPL exceeds 110 dB. A window at the passenger seat was fully opened. Same results can be observed with REW and LabView software. With the LabView software an additional filtering was performed, with two band-pass filters with frequency ranges from 1 Hz to 20 Hz and from 20 Hz to 40 Hz. Fig.3. Data processing block diagram in LabView program. Bartolj et al.: LOW FREQUENCEY NOISE MEASUREMENT IN THE PASSENGER CABIN 229 AAAA – 2023 – IZOLA - Conference Proceedings Fig.5. SPL level during the measurement, no frequency Fig.7. Closed window measurement without frequency weighting applied. REW software. weighting. REW software. Comparing the measurements, SPL difference of 15 dB can At an approximately equal speed of 110 km/h with the be observed when driving with speeds higher than w indow open, the SPL increased to around 112 dB, and 70 km/h on a highway. the duration of the open window was approximately 39 seconds, see Fig. 9 and Fig. 10. With the window open, the SPL increased by approximately 9 dB due to changes in air circulation, consequently leading to an increase in low- frequency noise. Fig.6. The SPL during the whole measurement session: no filtering (red), band-pass filtering 1 Hz to 20 Hz (green), band-pass filtering 20 Hz to 40 Hz (orange), and car speed in km/h (blue). LabVIEW software. Fig.8. Closed window measurement: no filtering (red), When driving on the highway at approximately 110 km/h, band-pass filtering 1 Hz to 20 Hz (green), band-pass the SPL with the window closed was around 103 dB, and filtering 20 Hz to 40 Hz (orange), and car speed in km/h the duration of the closed window was 2 minutes and 15 (blue). LabVIEW software. seconds. The cabin noise could be reduced with better sound insulation of the vehicle, improved window and A similar study was conducted involving the observation door sealing materials, enhanced suspension since of noise level in passenger car, but they used different car vibrations contribute to low-frequency noise, and a models a small opening of the window [20]. smoother driving style. As evident from the Fig. 8 and Fig. 10, the speed is In the Fig. 8 and Fig. 10 it is evident that with the window changing, yet the noise remains approximately the same closed, the SPL of the low pass filter up to 20 Hz is at the both with the window closed and open. It can be stated same level as the SPL without the filter, indicating that that the noise in the car cabin at speeds of 80 km/h and frequencies up to 20 Hz were dominant. above changes less than at a low speed. Bartolj et al.: LOW FREQUENCEY NOISE MEASUREMENT IN THE PASSENGER CABIN 230 AAAA – 2023 – IZOLA - Conference Proceedings A short summary of the findings is collected in the Table 2 and the Table 3 where SPL comparison with different frequency weightings and with open and closed window from LabView and REW programs are shown. SPL / dB Window closed Window open average max average max Wideband 101 110 113 117 1 Hz – 20 Hz 100 109 112 115 Fig.9. Open window measurement without frequency 20 Hz – 40 Hz 88 95 105 110 weighting. REW software. Table 3. SPL comparison with different frequency When comparing the closed and open window spectrum weightings, open and closed window. LabView software. measurements, see Fig. 11, it is evident that the difference in SPL at frequencies up to 20 Hz is The average and maximum SPL are recorded. As approximately 15 dB. With the window closed, higher mentioned earlier the driving conditions with open and harmonic components appear at the fundamental closed window are not exactly the same. frequency of 40 Hz, at around 80 Hz and around 160 Hz, likely due to resonances in the car's cabin. Fig.11. Comparison of open and closed window spectrum: open window (blue), closed window (brown). REW software. Comparison of the measured values using LabView and Fig.10. Open window measurement: no filtering (red), REW software shows good agreement, see Table 2 and band-pass filtering 1 Hz to 20 Hz (green), band-pass Table 3. filtering 20 Hz to 40 Hz (orange), and car speed in km/h (blue). LabVIEW software. 3. CONCLUSION With the window open, there is an increase in SPL around We investigated the low-frequency noise in the car cabin 700 Hz up to approximately 2 kHz. It could be due to the with the window either closed or open. The results absence of a windscreen on the microphone. indicate that opening the window affects the level of low- frequency noise in the vehicle's interior, especially in the SPL / dB Window closed Window open infrasound range, below 20 Hz. However, an open window average max average max also presents different challenges associated with wind, Wideband 100 111 113 117 affecting passenger comfort and health. The sound pressure levels of more than 100 dB and 110 dB in a car Table 2. SPL comparison with open and closed window. cabin with closed and open window respectively were REW software. measured. The noise level could be lowered with a Bartolj et al.: LOW FREQUENCEY NOISE MEASUREMENT IN THE PASSENGER CABIN 231 AAAA – 2023 – IZOLA - Conference Proceedings different vehicle design, alternative materials, and better [8] Meritve hrupa – Epi Spektrum d.o.o. Available: vehicle insulation. It is crucial to emphasize that our https://www.epi-spektrum.si/sl/varstvo- okolja/varstvo-okolja-pred-hrupom/splosno-o- findings are not easily replicable, as results vary ocenjevanju-hrupa-z-meritvami-2/ depending on weather conditions, traffic, and other [9] I. CORPORATIVA, La contaminación acústica, cómo factors. For quieter, more comfortable, and pleasant reducir el impacto de una amenaza invisible, journeys, collaboration between the industry, Iberdrola. Available: researchers, and designers is necessary to achieve https://www.iberdrola.com/sustainability/what-is- progress in this demanding goal. noise-pollution-causes-effects-solutions [10] NVH testing (noise, vibration, harshness), ATESTEO. Available: https://www.atesteo.com/en/testing/nvh-testing/ 4. REFERENCES [11] NVH Testing – Noise, Vibration, and Harshness, Data Acquisition | Test and Measurement Solutions. [1] B. Berglund, P. Hassmén, and R. F. S. Job, Sources Available: https://dewesoft.com/applications/nvh- and effects of low-frequency noise, The Journal of testing the Acoustical Society of America, vol. 99, no. 5, pp. [12] Noise, vibration, and harshness, Wikipedia. 2985–3002, May 1996, doi: 10.1121/1.414863. Available: [2] N. Broner, The effects of low frequency noise on https://en.wikipedia.org/w/index.php?title=Noise, people—A review, Journal of Sound and Vibration, _vibration,_and_harshness&oldid=1162539682 vol. 58, no. 4, pp. 483–500, Jun. 1978, doi: [13] Brüel & Kjær Hand-held Analyzer Types 2250 and 10.1016/0022-460X(78)90354-1. 2270 User Manual, Available: [3] C. S. Pedersen, Human hearing at low frequencies https://www.bksv.com/downloads/2250/be1713.p with focus on noise complaints, National Wind df Watch. Available: https://docs.wind- [14] GPS Visualizer. https://www.gpsvisualizer.com/ watch.org/Pedersen-human-hearing-low- [15] Edit et al. , New NMEA - GPS library, Jul. 17, 2015. frequencies.pdf Available: https://forums.ni.com/t5/Example- [4] M. Weichenberger et al. , Brief bursts of infrasound Code/New-NMEA-GPS-library/ta-p/3518883 may improve cognitive function – An fMRI study, [16] What is LabVIEW? Graphical Programming for Test Hearing Research, vol. 328, pp. 87–93, Oct. 2015, & Measurement. Available: doi: 10.1016/j.heares.2015.08.001. https://www.ni.com/sl-si/shop/labview.html [5] H. Moller and C. S. Pedersen, Hearing at low and [17] J. Mulcahy, REW - Room EQ Wizard Room Acoustics infrasonic frequencies, Noise and Health, vol. 6, no. Software. Available: 23, p. 37, Jan. 2004. https://www.roomeqwizard.com [6] D. L. Johnson, The Effects of High Level Infrasound, [18] Home, Audacity ®, Apr. 25, 2023. Available: Defense Technical Information Center, Fort Belvoir, https://www.audacityteam.org VA, Feb. 1980. doi: 10.21236/ADA081792. [19] VB-Audio Virtual Apps. Available: https://vb- [7] C. S. Harris, H. C. Sommer, and D. L. Johnson, audio.com/Cable/ Review of the effects of infrasound on man, Aviat [20] S. D. Haddad, An Investigation into the Significance Space Environ Med, vol. 47, no. 4, pp. 430–434, Apr. of Infrasound in some Passenger Cars, SAE 1976. International, Warrendale, PA, SAE Technical Paper 891158, May 1989. doi: 10.4271/891158. Bartolj et al.: LOW FREQUENCEY NOISE MEASUREMENT IN THE PASSENGER CABIN 232 MEASUREMENT AND CHARACTERIZATION OF CONTROL VALVES NOISE Egon Susič, Miha Pogačar Danfoss Trata d.o.o. Jožeta Jame 16, 1210 Ljubljana, Slovenia Abstract: Valves that control water flow in heating/cooling systems are often source of annoying noise. At the same time emitted noise can carry valuable information about flow condition. In spite that design of globe valves utilizes similar principle for reducing flow, the resulting noise exhibits many different characters and levels regarding dimension and material used. To enable noise control a measuring system is designed and applied predominantly at development stage of new valves. Most of typical noises are characterized by power spectral density function. By analysing noise and correlations with respect to internal geometrical features and flow conditions we manage to produce quieter valves and push the boundaries of operating conditions. Keywords: Control valve, noise measurement, flow induced noise 1. INTRODUCTION During the development process of valves some crucial constrains should be addressed. EU ESG policy demands Motorized Control Valve is a key element that delivers sustainable energy sources, low CO2 footprint to name water flow through heating/cooling systems is the main only few of them among many other directives [3,4]. subject of our research. Heating/cooling systems mostly Industry must follow these trends and demands as for utilize water to transfer heat energy from source to the example a usage of low temperature heating systems. user. Addressed systems can be regarded from different Consequently, lower temperatures mean smaller prospectives starting from district heating /cooling temperature differences and higher flow needed to networks down to local single flat HVAC system [1]. Many transfer adequate energy demands. Higher flow produces of these applications are designed to be located close to higher flow speed, and also reduction of material usage or directly in living spaces. During the valve operation, and smaller dimensions additionally increase flow speed. where due to water flow through connecting pipes and High flow speed inevitable produce higher flow noise valve internal cavities, audible noise might appear. which in some cases is very unpredictable in can develop When valve noise, also known as flow noise [2], becomes in tonal high frequency components that can be extra significantly higher than ambient noise, it can be a annoying and non-acceptable in living places. Upper limit nuisance factor that negatively affects people who live or is when cavitation takes place [5] and destructive abrasive work nearby. To tackle this issue, we built a measurement process deteriorate valve internal components and might system enabling us to monitor and examine valve end by destroy valve with unpredictable consequences performance also with respect to emitted noise. with high stakes. Furthermore, the noise can be a carrier of valuable Focusing on valve performance we should parametrize information reflecting typical and atypical flow behaviour. the process of delivering demanded flow. For this By analysis of this information and establishing purpose, we must address all known and measurable correlations between noise character and valve internal parameters that are present. First of them is the geometrical features we can influence valve design and differential pressure across the valve [6]. Next, high control noise level to certain extent. pressure is measured in front and static pressure behind Susič et al,: Measurement and Characterization of Control Valves Noise 233 AAAA – 2023 – IZOLA - Conference Proceedings the valve as well as water temperature. Valve opening stem. For the noise measurement there are three position is measured and driven by actuator which different positions on the piping system. produces enough force for closing the valve. Noise is Microphone MIC A is located inside semi-anechoic room measured independently in real-time. and measure structure borne emitted noise from radiator. This paper is structured as follows. Section 2. represents The radiator resembles a standard piece of room measuring system and testing principle. In section 3. are equipment that is usually located in living spaces. Radiator obtained results and analysis. Section 4. is discussion and pipe structure is connected to main hydraulic circle and the last section is conclusion. represents a waveguide of emitted noise from the control valve indicated as valve sample on Fig 1. 2. MEASUREMENT SETUP MIC B is positioned in the insulated box. Through this box Our measurement system is comprised of sensors that a steel pipe connected directly to outlet of valve (sample) simultaneously measure all above mentioned parameters. is mounted. By this arrangement we can measure emitted All sensors are calibrated and compliant to EN17025 noise presumably guided by the water and pipe itself. This standard demands [6]. The setup consists of pipes of is also regarded as structure noise. adequate diameter, water tank, pump and flowmeter. MIC C is located in valve’s proximity and measures valve This represents basic closed hydraulic loop. The water emitted noise as airborne and by changing a distance flow through this loop is controlled by multistage high- between source and microphone we can determine noise pressure pump driven by frequency-controlled motor and source taking into consideration inverse-proportional law the pressure differential across the valve is maintained by [8]. On another position far from the source the obtained a custom controller with variable closed-loop PID signal is regarded as background noise. algorithm. This loop resembles typical control valve Measurement protocol follows demand of control valve application environment. According to the relevant measurements defined in relevant standards while noise standard [7] valve’s characteristic is measured at constant measurements comply with adequate acoustic standard pressure drop over the valve while the opening position is [9]. independent control variable starting from closed position Hydraulic System control and data acquisition is to completely open. Valve opening position is measured presented in Fig. 2. Hydraulic closed loop is controlled by by incremental device that is side mounted to actuator standard PLC controller which is in feedback loop with and actuated by cantilever on clutch attached to valve control variable. Pressure drop over valve indicated as Fig.1. Measurement system with hydraulic closed loop control. Susič et al,: Measurement and Characterization of Control Valves Noise 234 AAAA – 2023 – IZOLA - Conference Proceedings Fig.2. System control and data acquisition setup pV in Fig .1. is predominantly used as constant value or we measure noise via MIC A, MIC B or/and MIC C linear increasing-decreasing function. All physical respectively. For this purpose, we utilize Dynamic Signal parameters mentioned in introduction section of this Analyzer (DSA in Fig. 2). Through software application paper are indicated in Fig,1 and Fig.2 respectively. based on LabView platform noise analysis is done. For Sensory signals are fed to A/D converter and processed each channel we use FFT procedure for obtaining power [10], filtered and stored every 200 ms. Graphs can be seen spectral, density shown on Fig .4, and 1/3 Octave analysis in Fig.3. This gives us resolution for monitoring and [10] that is compliant to standard ISO 3743. To have recording important parameters approximately 5 times complete picture of power spectra we plot waterfall graph faster that significant change of values occurs. In parallel in respect to control parameter that is indicated in Fig.5. 1400 70 1200 60 A)]B( 1000 50 vel [d ] e leis 800 40 o [L/h N te ra a], 600 30 [kP Flow a p 400 Flowrate [l/h] 20 ], d[N Force [N] 200 ΔpV [kPa] 10 Force MIC A [dB(A)] MIC C [dB(A)] 0 0 0 0,5 1 1,5 2 2,5 3 3,5 4 4,5 Stroke [mm] Fig.3. Measured quantities during functional testing of valves. Susič et al,: Measurement and Characterization of Control Valves Noise 235 AAAA – 2023 – IZOLA - Conference Proceedings Power Spectral Dencity 60 Sharp Flow Noise Structural with harmonics Tonal noise 50 40 B] 30 e [dd litup 20 Am 10 0 -10 100 1000 10000 Frequency [Hz] Fig.4. Power spectral density of Sharp, Structural with Harmonics and Tonal noise. 3. RESULTS The speed of actuator is small enough that we can Measurement protocol for valve characteristic assumes consider this process as quasi-stationary. Every 0,006 mm that recording is triggered at valve closed position and of stem displacement one record of all parameters and applied nominal force by actuator. Then the force noise pressure level are captured simultaneously. By decreases as actuator drives the valve stem to fully open doing so, obtained values can be regarded as continuous position and after in reverse direction towards closed curves with linear interpolation between consecutive position. Stem movement is indicated as stroke and discrete data points in respect to stroke domain. In Fig 3. represent independent control parameter. a typical measurement is shown. Additionally, to mentioned recording also power spectral density is calculated through FFT function and plotted into waterfall spectrogram shown on Fig. 5. Complete described measurement protocol runs in on-line mode with 200ms pace rate. 4. DISCUSSION Results of analysed measurements show that noise pressure level values slowly increase with increasing water speed through valve. During one measurement cycle we can accurately observe evolution of noise level. If these levels are within acceptance criteria, then valve pass quality check. However, beside level also noise character is evolving and this gives us a huge increase in complexity of noise emissions. Namely, characterization is Fig. 5. Waterfall spectrogram of atypical valve noise with done by power spectral density function which can respect to stroke position. discriminate among different phenomena that can take Susič et al,: Measurement and Characterization of Control Valves Noise 236 AAAA – 2023 – IZOLA - Conference Proceedings place in running water flow. Starting at very low flow at in Fig.4. This is considered as fluid induced noise and a almost close position of the valve we can detect a root cause is not well understood. Conditions are transients followed by flow noise. This flow noise unpredictable and tonal noise appearance is within very increases with increasing flow and might end in cavitation narrow window of measured parameters. if extremes are reached with pressure conditions in the flow. However, in some cases additional phenomena can be observed. These occurrences can be classified as noise 5. CONCLUSIONS due to structural vibrations and flow induced tonal effects. Vibrations are usually a consequence of design Valve noise can be annoying in cases where people are features of internal parts and cavities [11]. A significant exposed when appear in living spaces or their proximity. contribution can be from connecting pipes and other Valve producers must comply with legislation based on features in closed loop water circle. To minimize pumps adequate environmental standards and directives. Noise and other disturbances a rubber hoses (Fig.1.) are used to level is also measured to determine key feature of acoustically decouple valve from other elements. products quality. To reduce noise emissions a better valve Assuming that measured noise is prevalently from valve design is needed. It is well known that emitted noise we consider noise due to structural vibrations when contains valuable information about basic physical power spectral density shows natural frequency phenomena. For this reason, we utilize power spectral component and higher harmonics as well. These noises density function to characterize and differentiate among can be heard by human ear and recognised as kind of different noises. pattern. Many of different pattern might occur and we Proposed measurement setup works as an on-line clustered them in some main groups and addressed them monitoring and diagnostic system. Most of flow induced like rattling, fluttering, hammering, … On another hand noise phenomena can be related to valve geometrical tonal transients called whistling are represented by a features. By changing addressed features we can analyse single frequency component usually occurred at valve noise and with the gathered information produce frequencies higher than 1 kHz as can be seen better performing and quieter valves while preserving all desired features and functions. 4. REFERENCES [5.] EN 60534-2-1:2011. Industrial-process control valves – Part 2-1: Flow capacity – Sizing equations for fluid [1.] Rezaie, B., Rosen, M.A. District heating and cooling: flow under installed conditions, International Review of technology and potential Electrotechnical Commission, 2011 enhancements, Applied Energy, 93, pp.2–10, 2012 [6.] EN ISO/IEC 17025:2017. General requirements for [2.] Ng, K. Control Valve Noise, in ISA Transactions, 33(3), the competence of testing and calibration pp.275-286, 1994 laboratories, International Electrotechnical [3.] EU Technical Expert Group (EU TEG). Overview of the Commission, 2017 TEG final report on EU climate benchmarks and [7.] EN 60534-1:2005. Industrial-process control valves – benchmarks’ ESG disclosures - 30 September 2019. Part 1: Control valve terminology and general Available at: considerations, International Electrotechnical https://commission.europa.eu/system/files/2019- Commission, 2005 10/190930-sustainable-finance-teg-final-report- [8.] Kinsler, L.E., Frey, A.R. Fundamentals of acoustics, overview-climate-benchmarks-and-John Wiley & Sons, 1982 disclosures_en.pdf#:~:text=This%20assessment%20 [9.] EN ISO 3745:2012 Acoustics - Determination of must%20gradually%20integrate%20Scope%203%20 sound power levels and sound energy levels of noise emissions,for%20EU%20CTBs%20and%2050%25%20 sources using sound pressure – Precision methods for%20EU%20PABs. (Accessed 30 August 2023) for anechoic rooms and hemi-anechoic rooms, [4.] Connolly D., Lund H., Mathiesen B.V., Werner S., International Standards Organization, 2012 Möller B., Persson U., Boermans T., Trier D., [10.] Oppenheim, A.W., Schafer R.W. Discrete-time Ostregaard P.A., Nielsen S. Heat Roadmap Europe: Signal Processing. Pearson, 2010. Combining district heating with heat savings to [11.] Frank Fahy, Foundations of Engineering Acoustics, decarbonize the EU energy system, Energy Policy, 65, Academic Press, 2001 pp.475-489, 2014 Susič et al,: Measurement and Characterization of Control Valves Noise 237 Keynote Invited speech Modelling of sound and vibration using a virtual-source approach Prof. Dr. Goran Pavić Professor Emeritus at Institut National des Sciences Appliquées de Lyon, France E-mail: goran.pavic@insa-lyon.fr Goran Pavić holds BSC in Mechanical Engineering and PhD in Goran Pavić holds BSC in mechanical engineering and PhD in vibration and acoustics. After spending most of his career in industry research he joined in 1998 the National Institute of Science in Lyon as a full professor and, after the retirement, occupies the post of prof emeritus. His research covers sound and vibration energy flow, vibroacoustic modelling, characterization of sources and advanced experimental methods. He was actively involved in a large number of national and international research projects as well as in the organisation of numerous scientific events. The virtual-source approach enables the use of known analytical solutions of either a vibration or a sound field of a simple surrogate space to construct the field of a more complex space of arbitrary shape and boundaries. The field of the targeted space driven by the original primary excitation is obtained by superposing two fields of the surrogate space: one produced by the same primary excitation, the other by a (large) number of virtual sources. The latter are tuned to the primary excitation in such a way as to reproduce the required boundary conditions of the target space. The tuning is normally obtained by an inverse procedure which leads to the discretisation of boundary conditions. It is shown how the error induced by discretisation is evaluated and controlled. Numerous examples of vibration and sound fields are shown to illustrate the approach. 238 Contributed papers Soundscape and sound reproduction techniques 1. Application of Ambisonics to Building Acoustics – Challenges and Opportunities Armin Wilfling (IBO) 2. A Comparison between Real and Reproduced Sound Fields for Impact Noise Annoyance Ratings Martina Vrhovnik (InnoRenew CoE) 239 APPLICATION OF AMBISONICS TO BUILDING ACOUSTICS – CHALLENGES AND OPPORTUNITIES Armin Wilfling, Franz Dolezal IBO – Austrian Institute for Building and Ecology, 1090 Vienna, Austria Abstract: Concentrating on existing, standardized methods with an insufficient relation to human perception is not a satisfactory approach to comprehensibly characterize the nuisance caused by impact noise sources. Perception can no longer be condensed to a single number (the so called weighted impact sound pressure level), which is based on a low structural diversity of building components. Thanks to novel recording technologies, their unique combination and mathematical modelling, together with advanced methods for the analysis and interpretation of human perception, a holistic understanding of the acoustic and vibrating interrelationship between building and occupant can be created. Recordings of impact noise caused by neighbours are carried out using different methodologies with the intention to capture a full sphere surround sound for reproduction to test persons in an anechoic chamber. Impact noise caused by walking on a ceiling, built according to the (Austrian) standards, usually is of very low level with a low frequency emphasis. That specific kind of signal turned out to be challenging for Ambisonics© recording; the eigennoise of the microphones used became a limiting factor in the recording and reproduction process. Single microphone techniques were compared to multichannel recording systems, focusing on their performance on low level, low frequency signals. Keywords: Ambisonics, Impact Noise, Recording, Signal to Noise, IRIS Project, Eigenmike 1. INTRODUCTION 2. PRELIMINARY CONSIDERATIONS Almost 25 years ago Scholl and Maysenhölder rose the 2.1. The Ambisonics Signal question “Is todays description of the impact sound behaviour of ceilings correct and adequate?”, and pointed Ambisonics© is a full-sphere surround sound format. out, that a tapping machine does not represent the most Unlike most common surround sound formats, signals are common source of impact noise – a walking person [1]. not discrete (every signal goes to its dedicated speaker on They stated that wooden ceilings sound different from a predefined position). Ambisonics signals carry encoded concrete ceilings. Moreover the transmission of the position information, representing the sound field in the tapping machines’ noise depends on the impedance of the so-called B-format. ceiling. Nevertheless, the tapping machine still is the most This B-format signal is decoded individually at the listeners widely used standardized source for impact sound monitor setup, providing best possible representation of measurements in Europe. the recorded signal at that very own setup. To compare the annoyance of different ceiling types it is Therefore, on one hand side, Ambisonics© reproduction is desirable to record the noise signals and play the signals perfectly scalable to diverse setups. On the other hand to a test person in rapid succession [2]. The widely used side, the exactly same representation at every setup reproduction tests with headphones do not represent the cannot be guaranteed, which is the case in discrete actual sound field in a room, therefore multichannel surround sound formats. [5] loudspeaker systems may offer a better representation of the actual sound in the room. [3,4]. 2.2. Impact Noise Having these findings in mind, the IRIS Project chose to record and reproduce different sound-samples of ceilings 2.2.1. Impact Noise Sources for listening tests in the surround sound format Ambisonics©. Impact noise sources in residential areas normally include signals like walking – barefoot or with shoes -, jumping, Wilfling et al.: Application of Ambisonics to Building Acoustics – Challenges and Opportunities 240 AAAA – 2023 – IZOLA - Conference Proceedings mechanical noises etc. All those sources reproduced by We decided to use every-day impact noises and agreed on any ceiling construction do show a significant emphasis on having a test person (70 kg) walking – barefoot and with low frequencies (below 100Hz) [6]. shoes on. Recordings should be captured at a common listening height of 50 to 150 cm (lying to standing), not 2.2.2. Normative Representation close to the ceiling. Jumping on the floor – which produced higher levels and therefore better signal-to- In Austria an intermediate floor in a multi-storey dwelling, noise ratio (SNR) – was not considered, since we failed to according to the Austrian building legislation OIB-RL5 [7], keep the levels steady when repeating measurements. For fulfils the requirement, if it transmits a weighted better comparability and to get valid LńT,w values for our standardized impact sound pressure level LńT,w of 48 dB ceilings under test, a tapping machine was added to the or less. LńT,w as a single number rating does not take the comparison and to all further measurements. low-frequency parts of the signal into consideration. The tapping machine did, of course, generate far higher Requirements across Europe differ significantly between levels, but the spectrum produced is obviously very 48 and 65 dB LńT,w [8]. different from a person walking. Figure 2 shows the Low-frequent signals – causing the majority of complaints frequency responses of the tested signals, matched at 1 about impact noise, particularly in timber buildings [9] – kHz. At 100 Hz (1/3 octave) the spectra differ by more than are rarely represented in the valid standards. However, 10 dB, diverging even further below 100 Hz. also Hagberg states in [10] that in low frequencies, prevailing acoustic theories are doubtful and the design has to be adapted to a new methodology. 3. RECORDING REQUIREMENTS 3.1. The Input Signal To recheck the assumptions above, that the signal to be represented in the Ambisonics© systems is going to be very soft and low-frequent, we recorded different signals at the test suite at Technical University of Vienna (TU Wien) to compare their input spectra. Fig.2. Comparison of input spectra produced on a wooden ceiling (levels matched at 1kHz) walking barefoot, with shoes and tapping machine Still - in absolute levels the tapping machine produced far higher outputs, ranging more than 20 dB above the walking noise. 3.2. Recording Techniques These specific low-frequent, very soft signals to be captured turned out to be a challenging task for the Fig.1. Test suite at TU Vienna with tested floor and recording equipment. flanking path decoupling with flexible interlayers There are two basic approaches to recording that demanding signal and representing it in Ambisonics©. The test suite has the dimensions 540 x 330 x 340 cm and The first one is capturing the recording with an consists of a mass timber structure with flexible shells on Ambisonics© microphone – instantly coding the captured the inside of the walls with double planking of gypsum sound field into B-format. The second approach is to plaster board to avoid noise from flanking transmission as capture with a single low-noise microphone and adding well as disturbing noise from the outside of the test suite. the sound field information in postproduction. Latter Additionally flanking transmission is damped with flexible procedure is supposed to improve SNR. interlayers (figure 1) which are adjusted to the weight of A short overview about those approaches and its the floor test specimen for an optimized absorption effect. advantages and disadvantages is shown in Table 1. Wilfling et al.: Application of Ambisonics to Building Acoustics – Challenges and Opportunities 241 AAAA – 2023 – IZOLA - Conference Proceedings Ambisonics© 32 Microphone Array (4th Order) + Captures movement and sound field information Encodes B-Format signal using its proprietary software Convenient to use - Needs extra microphone for (calibrated) level information Class 1 Standard measurement microphone + Supposedly better SNR simple Calibration process - Massive postprocessing needed, virtually producing movement and sound field Table 1. Advantages and disadvantages of the two possible approaches to Ambisonics© recording 3.3. Microphone Selection In the anechoic chamber at TGM Vienna we compared SNR of a standard measurement microphone (Norsonic Fig.4. Comparison of the recordings of the pink noise 1225 / 1209) to the Eigemike Ambisonics© microphone bursts (left 28 dB(A), right 48 dB(A) standard microphone array (figure 3) by recording pink noise bursts at 28 dB(A) capsule (blue) and a single capsule of the Eigenmike - representing a good ceiling - and 48 dB(A) for a mediocre em32 (red) one. The first comparison met the expectation that the single microphone provided higher SNR. The difference of about 4 dB SNR was smaller than expected though. It still seemed worthwhile to give the Ambisonics© microphone array another try, keeping in mind, it would capture the real and actual sound field in the room, rather than adding a virtual one. Furthermore 1st order Ambisonics – used in the lowest coding band – should improve the SNR for the crucial low frequencies. In case the SNR would not be sufficient in the microphone array, we decided to also include a low-noise single microphone in the next step. That should improve the SNR for the single microphone technique by about 8 dB. Experiments to denoise the recordings or to introduce noise-shaping-algorithms in postproduction were turned down, since that editing produced artefacts clearly audible. Trying a low-noise microphone and comparing it to the standard microphones we already tested, was the next Fig.3. Comparison of the eigennoise of Eigenmike em32 step - back in the test suite of TU Wien. To turn its Ambisonics microphone to a Norsonic 1225/1209 receiving room into a more living-room-like acoustic situation, we added absorption to the walls and floor, The SNR was captured and compared in those recordings, decreasing its reverberation time below 0.5 s. A Gras 26 shown in figure 4. HF / 40 EH low noise microphone was added to the test setup (figure 5). Wilfling et al.: Application of Ambisonics to Building Acoustics – Challenges and Opportunities 242 AAAA – 2023 – IZOLA - Conference Proceedings different. The localisation of the sound sources seemed to suffer a bit from the decreased number of capsules, but was in a satisfactory range. In contrast to the Eigenmike em32 the Ambeo VR does not generate a B-format file, thus it can be easily generated by a software provided by the manufacturer. Advantages of the Ambeo VR are that it provides four XLR outputs, allowing the user to pick from a wide variety of audio interfaces and its very competitive pricing. Ambeo VRs SNR may also benefit from using higher quality preamplifiers. For this test decent quality preamplifier (microphone amplifiers in the Roland Fig.5. Setup of the different recording approaches at the Octacapture and the MOTU 8pre interfaces) were used. adapted test suite A few other microphones we tested, but unfortunately they did not meet our expectations. By the time those tests were conducted, the Ambisonics© listening facility at InnoRenew was ready to start with the first listening tests (figure 6). 4. RECORDING PROCEDURE First listening tests indicated that the superior sound field information captured by the Ambisonics© microphone It was then decided to define a recording procedure for array outweighed the better SNR captured by the single recording Ambisonics sound samples of ceilings for the microphone. However, the relatively low SNR of the IRIS project. Those demands had to be meetable in the Eigenmike em32 compromises measurements and test suite as well as in in-situ recordings at building sites. reproduction – especially for ceilings that exceed Signals were defined to be a person walking on the floor normative demands. But to put that into the right above at 80 and 100 steps per minute (using a metronome perspective: walking on a floor which is far better than the for comparability), each with shoes on and barefoot. Our Austrian standard or building legislation requirement, is test-walker weighs 70 kg. As a further signal a tapping hardly audible in typical living environments [11]. machine was added. The signal is recorded in Ambisonics©, but primarily the tapping machine is needed to measure LńT,w of the specimen for documentation. Microphones are level matched (the actual level is calibrated for the LńT,w measurement) by playing pink noise at 70 dB(Z) on a Genelec 8020 loudspeaker. The same model used in the listening room. At least 3 positions - recording the whole set of signals - were postulated to represent different acoustic shadings within a listening room. Each sample recorded shall be 1 minute. Since more than one recording devices is used (Ambisonics©, standard measurement and vibration – which is not mentioned in this paper) some kind of slate was demanded to synchronize the recordings for postproduction and reproduction. We decided to jump at the start of each recording, giving a good impulse to easily synchronize. 5. CONCLUSIONS It turned out, that using Ambisonics© systems for building Fig.6. Listening test in the unechoic chamber at acoustic purposes requires a lot of preparatory work and InnoRenewCoE preliminary studies that have to be done. Comparison with standard and low noise microphones revealed a low Adding more options, a Sennheiser Ambeo VR – a four SNR that might be too low for a reproduction with test capsule Ambisonics© microphone –was investigated as persons simulating the real noise impression in the test well. The SNR was in a comparable range to the Eigenmike suite at InnoRenewCoE in Izola. It could be shown that em32, although the eigennoise did sound slightly standard measurement microphones as well as Wilfling et al.: Application of Ambisonics to Building Acoustics – Challenges and Opportunities 243 AAAA – 2023 – IZOLA - Conference Proceedings particularly low noise microphones do fulfil the necessary [3.] Wiersdorf, H., Raake, A., Spors, S. Psychoakustik der requirements but obviously cannot record the spatial Wellenfeldsynthese: Vor- und Nachteile binauraler impression required for the reproduction. At the present Simulation, in Proceedings of DAGA, Darmstadt 2012. stage of the research project, the majority of the foreseen [4.] Koehler, M., Weber, L., Späh, M. recordings have been carried out with all mentioned Trittschallminderung von austauschbaren equipment. Recordings are provided for the use in the test Bodenbelägen, Fraunhofer IRB Verlag, 2014. suite, where they are verified if they are suitable for our [5.] www.waves .com Ambisonics Explained: A Guide for purpose. If this is not the case, additional efforts of post Sound Engineers, 2017. Available at: processing will be necessary. https://www.waves.com/ambisonics-explained- guide-for-sound-engineers [6.] Maack, J., Möck, T. Trittschallschutz, in Bauphysik-6. ACKNOWLEDGEMENTS Kalender 2009, John Wiley & Sons, 2009. [7.] OIB Richtlinie 5, Schallschutz. Österreichisches The authors gratefully acknowledge the financial support Institut für Bautechnik 2019. of the Austrian Science Fund (FWF) research project I [8.] Rasmussen, B. Acoustic classification of buildings in 5503-N Engineered wood composites with enhanced Europe – Main characteristics of national schemes impact sound insulation performance to improve human for housing, schools, hospitals and office buildings, wellbeing. in Proceedings of the Baltic-Nordic Acoustics Meeting, 2018. [9.] Dolezal, F., Neusser, M., Teibinger, M., Bednar, T. 7. REFERENCES Akustik Center Austria – neue Prüf- und Forschungskompetenz für Holzkonstruktionen in Österreich mit Fokus auf tiefen Frequenzen, in [1.] Scholl, W., Maysenhölder, W. Wird das Proceedings of DAGA, Aachen 2016. Trittschallverhalten von Gebäudedecken derzeit [10.] Bard Hagberg, K. Management of acoustics in richtig und ausreichend beschrieben? in WKSB, lightweight structures, Dissertation, Department of 43/1999. Construction, Lund University, 2018. [2.] Bregman, A. Auditory Scene Analysis: The Perceptual [11.] ÖNORM B8115-5:2021 Schallschutz und Organization of Sound, MT Press, 1994. Raumakustik im Hochbau - Teil 5: Klassifizierung, Austrian Standards, 2021. Wilfling et al.: Application of Ambisonics to Building Acoustics – Challenges and Opportunities 244 A COMPARISON BETWEEN REAL AND REPRODUCED SOUND FIELDS FOR IMPACT NOISE ANNOYANCE RATINGS Martina Vrhovnik1, Dr. Rok Prislan1 1 InnoRenew CoE, Acoustics Laboratory, Izola, Slovenia Abstract: of seven horizontal levels, named alphabetically from top (A) to The most common way to evaluate subjectively perceived an- bottom (G). Two loudspeakers are mounted on the upper level noyance of impact noise is conducted with the help of listening A) and lower level G), while levels B) to F) are arranged in circular tests. Impact noise recordings are played back through headrings holding the loudspeakers. Their arrangement is horizon- phones or loudspeakers under controlled conditions in which the tally symmetrical with ring D) defining the plane of symmetry. reproduced impact noise should be perceived and rated consis- tently to the impact noise in a real environment. According to the standard ISO/TS 12913-2 binaural reproduction are preferred over a multichannel loudspeaker approach. Arguing that multichannel playback lacks a standard technique and examples of best practice. On the other hand, studies in the built environment imply that playback through loudspeakers contributes to the ecological validity of the test environment. Participants are able to move their head or entire body and get a better impression of the sound field which leads to more valid test results. As such further investigations related to the optimal listening approach are needed. This paper presents preliminary results of a listening test in which the participants rated their annoyance when exposed to real impact sources and their acousti- cal reconstruction in a direct comparison. Participants were exposed to real impact noise, excited on the upper floor, and its reproduction through a higher order Ambisonics system. Annoyance ratings for different impact noises were evaluated regard-Fig. 1. Symmetrical Ambisonics loudspeaker setup at the ing the distinguishability between real sound field and its repro-InnoRenew CoE. Number of loudspeakers per ring: Ring A duction through loudspeakers. Results expose differences in the & G: 2, B & F: 6, C & E: 12, D: 24 perceived annoyance depending on the reproduction technique. Keywords: higher order ambisonics, plausibility, qualitative test-The reproduction of auditory sound scenes using the Higher ing, impact sound Order Ambisonics (HOA) method differs from a ”standard” multi- channel loudspeaker setup. Unlike other multichannel formats, the reproduction of the sound field is loudspeaker independent. 1. INTRODUCTION The sound field is recorded with a spherical microphone and the In simu applications are used as a substitute for acoustic real- recordings are encoded in the so-called B-format. Each signal in ity in terms of comparative qualitative evaluation of a sound this format contains and reproduces specific sound field proper- scene. The InnoRenew CoE’s acoustics laboratory is equipped ties in the form of spherical harmonics and therefore takes into with a 64-channel Ambisonics loudspeaker system that serves account the directional properties [1]. as a playback environment for listening tests that we use to in- For a realistic reproduction with HOA, a so-called perceptual vestigate acoustic comfort in the built environment. The loud- ”sweet spot” at a central listening position is required [4]. In the speakers are mounted on a spherically arranged support system loudspeaker system presented here, this central listening posi- installed in the anechoic chamber of the laboratory. It consists tion is also the center of the sphere created by the loudspeaker 1st Author Surname et al.: Paper title 245 holding system itself. In the technical specifications of ISO 12913-2 it is argued that multichannel reproduction lacks of a standard technique, examples of best practice, and that there is still a need for research in realistic sound reproduction [2]. However, in addition to the commercial breakthrough in surround sound applications offered by Google 1 or Facebook 2 , B-format play- back has gained recognition in the scientific environment and is increasingly used as an audio framework for virtual reality [4]. One of the drawbacks of this method is its poor directionality, especially at first order [1]. In addition, high-frequency com- ponents reproduced in a high-density loudspeaker array can be degraded if decoded to a low- order of Ambisonics [3]. Impair- ment of localization is also possible if the number of loudspeak- ers exceeds the minimum requirement [3]. To assess a realistic reproduction, it is essential to diagnose and correct sensory deficiencies in the reproduction environment with HOA. Therefore, a direct comparison between a real sound scene and its repro- duction was necessary. Due to the construction of the acoustics laboratory and the anechoic chamber, it is not possible to create ”everyday” impact sound (e.g. walking, slamming doors etc...). We decided to set up the Ambisonics loudspeaker setup at a dif- ferent location where the generated impact sounds could reflect everyday acoustic conditions of a living environment. 2. METHOD Fig. 2. Listening test setup on the second floor of the In-2.1 Test Setup and Measurements noRenew CoE In one of the offices on the second floor of the InnoRenew CoE, we built a reduced version of the HOA consisting of 18 loud- speakers, also assembled in a spherical arrangement, and a sub- Ambisonics IEM plug-in suite developed at the University of Mu- woofer with a crossover frequency of 60 Hz. sic and Performing Arts in Graz, Austria 4 . The speaker setup consisted of four levels with 2 speakers The spherical recordings were then equalized and normal- on the top (plane A), 4 speakers on level B, 8 loudspeakers at lisized. This was done by measuring the sound pressure level of tening height at 1.3 m on level C, and 4 loudspeakers on the floor the playback with the omnidirectional microphone and adjust- (level D). The floor of the room was wooden parquet the ceiling ing it within octave bands until it matched the level of the real was composed of an sound absorbing suspended ceiling. To fur- impact sound. ther reduce room reflections, a wooden structure (3.5m x 3.5m) was built to support sound-absorbing curtains that surrounded 2.2 Listening Tests the listening test area. Impact noise was generated on the third floor, directly above the listening test area (see Fig.3) . The focus of our investigation is on evaluating the perceived differences between real and reproduced impact sound. For A standardized tapping machine and real barefoot walking this purpose a preliminary study has been performed from (pace: 90 bpm) were generated as impact sound. We recorded which the pilot listening tests results are presented. In a direct both source types using an omnidirectional microphone and comparison, nine participants rated the differences between a spherical microphone (mhacoustics’ Eigenmike em32). Both the real impact sound and its reproduced counterpart on nine were placed in the ”sweet spot”, at 1.3 m ear level. The repro- qualities. First, they answered a closed question ( ”yes” or ”no” ) duced impact sounds were played back in Reaper 3 . For decod-whether they perceived any difference at all. The ratings of the ing to the loudspeakers, we used the AllRAD decoder from the following seven qualities,( annoyance, plausibility, thumping 1 noise, tone color, localizability, clarity, and artifacts), included https://developers.google.com/vr/ios/ spatial-audio two questions. In the first question, participants rated the 2 https://www.facebook.com/groups/ difference between two sounds with respect to a particular ambisonictoolkit/ 3 https://www.reaper.fm/ 4 https://plugins.iem.at/ 1st Author Surname et al.: Paper title 246 inal scaled questions for the quality overall difference ( ”yes” or ”no” ) and which of the two sounds corresponded more to a particular adjective describing the quality ( 1st or 2nd). tapping machine noise 6 5 4 3 answer 'yes' real 2 'no' repr. 1 Fig. 3. Position of the tapping machine and walking path (pink) on the third floor of the InnoRenew CoE 0 ence er? property using a nine-point scale. The second question was also brigher? clear a closed question, asking which of the two sounds presented e thumping ence clarity e artifacts ( ”1st” or ”2nd” ) corresponded more to certain adjectives de-e annoying? e plausible? ence artifact scribing the qualities (see table 1). To query another perceived ence thumping mor ence tone color overall differ ence annoyance mor ence plausibility mor differ difference that was not mentioned, participants could describe ence localazability easier to localize? differ differ differ differ e noise-lik it in the last open question using the label ” other”. differ differ mor Quality Property Fig. 4. Listening test results for the impact sound source Annoyance more annoying ”tapping machine”. Plausibility more plausible Thumping noise more thumping For the tapping machine all participants were able to Tone color brighter distinguish between the real tapping sound and its reproduced counterpart. The real sound was rated as more annoying but Localizability easier to localize not as plausible as the reproduction. The real sound was also Clarity clearer rated as more thumping, easier to localize, and clearer. The Noise artifacts more noise-like artifacts reproduction was rated as having more noise like artifacts. In regard to tone color the averaged results show the participants’ Table 1. Evaluated qualities and their descriptive proper- indecision. On average, half of them said that the reproduction ties sounded brighter, while the other half found the real impact sound brighter. Since auditory memory is only a few seconds long, the par- ticipants needed to hear the impact sounds to be compared In terms of individual scale scores for each quality, partic- more often in order to answer all of the questions correctly. ipants’ average scores ranged from 3 to 4.5, with the highest Therefore, the two impact sounds were played consecutively for scores given for differences in thumping noise, tone color and the listeners to rate each of the assessed quality. After the play-clarity. back, the participants had 30 seconds to answer each question. The results for the impact sound type walking are quite different. On the one hand, for each quality, the standard deviation as well as the range of values given in terms of perceived difference 3. RESULTS are lower than for tapping machine. One participant perceived The boxplots 4 (tapping machine) and 5 (walking) show the av- no difference (represented as a dot) between the reproduced eraged results of nine participants for all 8 qualities. The odd and the real walking sound. Except for two participants, the dif- columns show, in chronological order, the responses to the nom- ference in annoyance was rated as slightly different on average 1st Author Surname et al.: Paper title 247 walking noise difference scores, this could indicate that the tonal impression was not accurately reproduced. In this respect, for future listen- 6 ing tests we envision the application of additional loudspeakers into the ceiling area. It is as well interesting to point out that 5 for the brightness, the listeners were highly confused/indecisive which of the stimuli is brighter, although they rated the differ- 4 ence in brightness relatively high. It will be interesting to investigate if this behavior will persist when a higher number of test 3 answer subjects will be evaluated. The reproduction of the barefoot walking sound was more 'yes' real 2 difficult for the participants to distinguish. Barefoot walking sound is characterized by its dominant sound energies at low fre- 'no' repr. 1 quencies below 100 Hz. We used a subwoofer to reproduce fre- quencies below 60Hz. This could be a reason for a more plausi- 0 ble reproduction compared to the reproduced tapping machine sound. The quality thumping noise has the highest average rat-ence er? ing among all participants. This is because apart from clarity brigher? clear e artifacts with a similarly high rating, all other ratings are considerably e annoying? e plausible? e thumping ence clarity ence artifact lower. This suggests that the reproduction could be improved ence thumping mor ence tone color by including the reproduction of impact vibration [5]. overall differ ence annoyance mor ence plausibility mor ence localazability easier to localize? differ differ However, one must also consider the composition of the differ differ differ differ e noise-lik differ mor two tapping sounds. While the tapping machine produces a con- tinuous hammering, walking produces more hollow, thumping noise at greater intervals between the actual impacts. The au- Fig. 5. Listening test results for the impact sound source thor assumes that because of their composition, the auditory ”walking”. analysis to determine differences was inherently ”easier” for the tapping machine. by the other participants. However, they did not have the same impression of which of the two sounds was ”more annoying” . 5. CONCLUSION On average, half of the participants chose one of the two sounds at a time. In this paper we presented a listening test procedure to evalu- Similar results are shown for ”more plausible” and ”clearer” . ate the distinguishability between real impact noise of a tapping On average, the reproduction was rated as slightly brighter, but machine and barefoot walking and its reproduced counterpart also as having more noise-like artifacts. The real walking sound in B-Format, decoded to an 18-channel loudspeaker system. The was rated as more thumping and easier to localize. presented preliminary results included the response obtained The differences between the real and reproduced walking were from the pilot listening tests. In a direct comparison, 9 partic- with one exception ( artifacts), not rated higher than with 4, ipants rated the differences using 9 auditory qualities. The re-which would be moderately different. On average, the ratings sults suggest that differences in tapping machine sound were for all qualities did not exceed 2.5. For both impact sound types easier to detect due to the lack of high-frequency content in the the qualities thumping noise and clarity had the highest differ-reproduction. Differences in walking sound were more difficult ence ratings on average across all participants. to perceive, possibly due to the plausible reproduction of the dominant low-frequency content of this impact source. Because differences in thumping noise had the highest ratings for both 4. DISCUSSION & OUTLOOK impact sources, inclusion of impact vibration in the reproduc- In terms of the results, it can be said that the reproduction of tion could result in a more plausible auditory simulation. both impact sound types, tapping machine and walking, is dis- It should be noted that the results of a pilot study were pre- tinguishable from their real counterpart. For each quality, the sented. As such, they are based on an insufficient number of par- ratings of differences for the tapping machine sounds have a ticipants to draw statistically significant conclusions. Neverthe- wider spread with maximum values up to 6 ( ”very different” ). less, some interesting correlations between the observed stim- One reason for this could be the insufficient reproduction uli were uncovered, which will lead to technical and question- of higher frequency content, which could lead to a lack of clar- naire improvements in further listening tests that will include a ity [6]. Since the qualities tone color and clarity have the highest much larger number of participants. 1st Author Surname et al.: Paper title 248 6. ACKNOWLEDGMENTS Soundscape - Part 2: Data collection and reporting require- ments, International Organisation for Standardization, 2018. The authors gratefully acknowledges the European Commission for funding the InnoRenew project (grant agreement #739574) [3] Shu-Nung Yao and Tim Collins and Peter Jančovič Timbral under the Horizon2020 Widespread-Teaming program and the and spatial fidelity improvement in ambisonics, Applied Republic of Slovenia (investment funding from the Republic of Acoustics, 93, pp.1-8, 2015. Slovenia and the European Union from the European Regional [4] Zotter, Franz and Frank, Matthias Ambisonics: A Practical Development Fund). The authors acknowledge the financial 3D Audio Theory for Recording, Studio Production, Sound support from the Slovenian Research Agency (research core Reinforcement, and Virtual Reality, Springer International funding No. Z1-4388, Toward better understanding the diffuse Publishing, 2019. sound field, and research core funding No. J4-3087, Engineered [5] Prislan, Rok The Design of an Impact Sound Vibration Expo- wood composites with enhanced impact sound insulation per- sure Device, in Proc. of the 10th Convention of the European formance to improve human well being). Acoustics Association, 2023. [6] McKenzie, Thomas and Murphy, Damian T. and Kearney, 7. REFERENCES Gavin Diffuse-Field Equalisation of Binaural Ambisonic Ren- [1] Arteaga, Daniel Introduction to Ambisonics, 2015. dering, Applied Sciences, 2018. [2] International Organisation for Standardization Acoustics - 1st Author Surname et al.: Paper title 249 Contributed papers Noise and vibrations 1. Proposing noise barriers along existing national roads through settlements - the case of Novo mesto and its surroundings Mihael Žiger (Nacionalni laboratorij za zdravje, okolje in hrano) 2. Measurement and modeling uncertainty in accredited acoustic procedures Antonio Petošić (University of Zagreb, Faculty of electrical Engineering and Computing) 3. Development of a special standard for outdoor music events in Slovenia based on measurements and calculations Luka Čurović (University of Ljubljana, Faculty of Mechanical Engineering) 250 PROPOSING NOISE BARRIERS ALONG EXISTING NATIONAL ROADS THROUGH SETTLEMENTS - THE CASE OF NOVO MESTO AND ITS SURROUNDINGS Mihael Žiger Nacionalni laboratorij za zdravje, okolje in hrano (NLZOH), Prvomajska ul. 1, SI – 2000 Maribor Abstract: Noise barriers are widely used in Slovenia along motorways and more recently along railway lines, to cope with exceedances of limit values, identified by noise monitoring and strategic noise mapping. Noise barriers along other national roads are mainly used for new developments (bypasses, reconstructions). As the use of passive measures along existing national roads through settlements was considered the only appropriate option, there has been virtually no systematic approach to noise barriers along the existing national roads. In 2021, the Directorate of the Republic of Slovenia for infrastructure (DRSI) commissioned noise studies along major roads under its management. Our institution has carried out a noise study for the area of the town Novo mesto and its surroundings, for three road sections with a total length of about 15 km, mostly within built-up areas. We have established baselines for the feasibility of noise barriers and found seven areas where noise barriers can be proposed. The results show that along the existing national roads through settlements - in addition to the usual passive measures on buildings, and considering limited possibilities of using quieter asphalt and lower speed limits - it is reasonable to consider noise abatement also by noise barriers. Keywords: noise, barrier, road, settlement, study, Novo mesto, Ribnica 1. INTRODUCTION model calculations and measurements that we carried out, which were of course necessary for such a task. This article presents findings of the noise study for three The paper is structured by first providing baselines for the road sections with a total length of about 15 km in the feasibility of noise barriers, then roughly presenting area of the town Novo mesto and its surroundings, model calculations including road, traffic and commissioned by the Directorate of the Republic of environmental data, and finally, after identifying suitable Slovenia for infrastructure (DRSI) and carried out by areas, moving on to the noise barrier proposal. NLZOH [1]. The road sections are: R2-419/1203 Soteska - Novo mesto, km 12,253 - 14,513, G2-105/0256 Novo mesto (Revoz) – Metlika, km 0 - 3,184 2. BASELINES FOR NOISE BARRIER PLACEMENT and G2-106/0263 Žlebič – Kočevje, km 0 to km 9,505. The work began by establishing appropriate baselines for The road sections are in the area of the municipalities the placement of noise barriers, a result of joint work of Ribnica and Novo mesto, mostly running through built-up noise companies proposing barriers at different areas, the areas. The main goal of the study was identifying areas to engineer and the commissioner: be protected by noise barriers and proposing them. 1. The noise barrier should protect at least a group of In this article, we focus on identifying areas, suitable for three overexposed (exposed to noise, which exceeds the noise barriers, the criteria for their placement and the limit values) residential buildings and should have no solutions for barriers at the study level, rather than on the interruptions (pedestrian passageways are allowed with appropriate overlaps or doors). Žiger: Proposing noise barriers along existing national roads through settlements – the case of Novo mesto and its surroundings 251 AAAA – 2023 – IZOLA - Conference Proceedings 2. When protecting unbuilt areas, only public areas such the CadnaA software tool (DataKustik, Munich, Germany, as the surroundings of educational establishments version 2022). The calculation method for road noise was (kindergartens, schools), cemeteries and, conditionally, XPS 31-133, and we have introduced road, buildings, park areas and areas adjacent to multi-apartment terrain, ground absorption, existing (noise barriers), all buildings should be considered. All such areas should be with the appropriate properties. We calculated noise subject to a detailed inspection to verify the justification maps at a height of 4 m and maximum noise levels at all and appropriateness of the protection. noise-sensitive buildings up to a distance of at least 200 m 3. Before proposing a noise barrier, it is also necessary to from the road. Calculations were carried out for all the check the intended land use; barriers (for the protection mandatory noise indicators (Lday, Levening, Lnight, and Ldvn). of living space) should be mainly planned for residential The noise maps were calculated at a 5 m x 5 m grid. In the areas or areas with predominantly residential use. Central areas where noise barriers were proposed (see next (mixed) areas and areas with scattered buildings should section), further calculations were carried out, such as a usually not be protected by barriers. noise map at a height of 2 m, façade maps for every 4. The minimum height of the barriers shall be 2.0 m. The building floor, noise at the receivers on the most exposed recommended height is between 2.0 m and 2.5 m, façades per floor, all with and without the proposed noise although higher barriers are possible in exceptional cases barriers. (which must be justified in the proposal). The horizontal and vertical alignment of the roads in the 5. The noise barrier shall reduce the noise by at least 5 dBA model were taken from the operational monitoring [2], at the most exposed assessment point (at a height of 2 m) with minor corrections only. Road data were also taken and provide a sufficient (best possible) economic eligibility from operational monitoring – traffic speeds, asphalts, index (index defined in [3]). traffic flow, but with field verification. Traffic on most 6. The proposal of noise barriers may also include a roads was also based on traffic from operational removal of a connection to a local road or driveway if monitoring, increased by an annual growth rate of 1.5% access is provided otherwise. (after analysis of traffic counter data) until 2045. On the Regarding the placement of shorter barriers by the section G2-105/0256 Novo mesto (Revoz) - Metlika, which owners themselves to protect their own buildings or areas is planned to be bypassed by a parallel road, such an from noise (such barriers were not the subject of the increase in traffic is not realistic and we took into account study), we have proposed that such barriers 1. shall not the relevant traffic study of this bypass. compromise traffic safety, 2. shall comply with the Buildings were taken into account according to the requirements of the municipal spatial plan, 3. shall be operational monitoring [2]. However, the buildings in the sufficiently absorbent at the road side (noise absorption first line of houses along the road were inspected by field panels that reduce noise by at least 8 dB on reflection, visits in autumn 2021 and modified, added or removed as conditional gabions - exceptions to this condition being necessary. The configuration of the terrain in the form of only possible if there are no existing or planned noise- contour lines was taken from the operational monitoring, sensitive buildings or areas on the opposite side of the with minor corrections. The ground absorption in the road, and 4. shall be sufficiently insulating to reduce noise operational monitoring had been set according to the by at least 25 dB on transmission (noise barriers, gabions, actual use and was not presented in a way that would solid walls, etc. ) - gaps in the barrier (between the posts uniquely allow its reproduction, so in this study it was set and the panels, or between the base and the panels) or according to the planned land use defined in the plans of other openings or breaks shall be avoided, as they the municipalities (Ribnica and Novo mesto). For example, significantly reduce the noise-abatement effect of the transport infrastructure (road) and water areas were barrier. considered as reflective (G = 0.0), most of the natural and agricultural environment as absorptive (G = 1.0) and most of the built-up areas as partially absorptive (G = 0.7). For 3. CALCULATION OF ROAD NOISE residential and other noise sensitive buildings, the national noise limit values (III. noise protection zone) Lday We set up a three-dimensional noise model. The 65 dBA, Levening 60 dBA, Lnight 55 dBA and Ldvn 65 dBA for the modelling and noise calculations were carried out using road as a noise source were used. Other noise sources Žiger: Proposing noise barriers along existing national roads through settlements – the case of Novo mesto and its surroundings 252 AAAA – 2023 – IZOLA - Conference Proceedings were not taken into account, as they were negligible Pie in Pina Mlakarja) has no entrances from the main road compared to the roads under consideration. and thus a noise barrier is possible. The area is fenced even in its existing state, the house plots are mostly above the main road and therefore concrete retaining walls of 0 4. IDENTIFYING AREAS SUITABLE FOR NOISE BARRIERS to 1.8 m height are present, on top of which are also fences of different heights and materials, some of them The descriptions and tables start with the road section R2- being even full, but without defined acoustical properties. 419/1203 Soteska - Novo mesto from km 12,253 to km A suitable noise barrier in area 3 should be somewhat 14,513, then G2-105/0256 Novo mesto (Revoz) - Metlika higher than in other areas, due to existing height from km 0,000 to km 3,184, and finally G2-106/0263 differences. Žlebič - Kočevje from km 0,000 to km 9,505, all in the Area 4 (see figure 4), section 0263: The settlement of Grič direction of stationing. “Overexposed” in this chapter has a compact built-up area to the left of the road and means exposed to noise above the limit value Lnight 55 dBA extends from km 1.2 to 1.8. The planned land use of the in 2045. area with the houses is SS - residential. For the first two Areas suitable for noise barriers are presented in the buildings (Road I 10, Road I 12, km 1.24 to the left), following text and figures from 1 to 6. The road sections comprehensive noise abatement is not possible due to are shown in red, noise sensitive buildings in blue, other driveway and proximity. The first two buildings are buildings in grey and the proposed noise barriers in black. followed by an area accessed by local roads (to Road IX at Area 1 (see figure 1), section 1203: Between km 12.3 and km 1.31, Road VIII at km 1.38, Road VII at km 1.46 and 12.4 there are three overexposed residential buildings Road VI at km 1.52) up to the building Road VI 2 at km and their functional land to the right (Novo mesto: Drska 1.55. In this area, the road reconstruction project foresees 2, Drska 1 and Drska 24), which are mostly slightly below the removal of the connections of these four local roads the road in terms of elevation and do not have direct to the main road, which allows the planning of a noise access from the national road. There are good possibilities barrier in this area. From including the building Road VI 2 for a noise barrier. The beginning of the barrier is limited at km 1.55 (as well as in the whole area 4 to the right of by the intersection with Slavka Gruma Street, the end by the road), we again have individual entrances to the main the cycle path and driveway at Drska 24. The regional road road, which make noise barriers impossible. is cut in immediately after the cycle track, so from the Area 5, section 0263 (see figure 5): To the left of the road cycle track onwards we also have noise protection of the is the area of the primary school in Ribnica, extending area due to the road cut. The planned use of the area of from km 3.66 to the junction with Kolodvorska Street at the buildings is SS - residential. km 3.74. There are two tall school buildings, named C and Area 2 (see figure 2), section 0256: Between km 0.6 and A, along the road, and it would not make sense to shield 0.9 to the left there is a residential area along Ulica Ivana the buildings by a noise barrier. However, we consider this Roba. The first two buildings are above a road cut (Nos 33 public area worth protecting by a noise barrier. In front of and 2), and then the road passes into the embankment, so Building A, approx. km 3.71 to 3.73, there is a playground that between km 0.7 and 0.85 the buildings are below the with play equipment for small children, and road in height. The entire area of houses is accessed from schoolchildren occasionally gather between buildings C a junction at km 0.9, so there are no road accesses from and A (in front of the entrances to both school buildings). the main road, the only pedestrian access being the The school's land use is CD - central. Building A is under staircase at No 33. There are more than 10 overexposed monument protection, and there are two wooden residential buildings. The planned land use of the area sculptures in front of the school. There is already a fence with the houses is SS - residential. on the same line where the noise barrier would be Area 3 (see figure 3), section 0256: In the area from km located. The existing fence has already been demolished 1.5 to 1.95 to the right there is a larger built-up area along in its western part (probably for access to the parking Pie in Pina Mlakarja Street. The initial (3) and final (5) few spaces at the school) and we are not proposing a noise buildings (with the address Belokranjska road) have barrier in this part as it is not necessary to protect the play individual entrances or accesses from the main road, areas and entrances to the school buildings. The terrain, while the central part (10 houses with the address Ulica including the road, is predominantly flat. Žiger: Proposing noise barriers along existing national roads through settlements – the case of Novo mesto and its surroundings 253 AAAA – 2023 – IZOLA - Conference Proceedings Area 6 (see figure 5): To the right of the road in Ribnica, 5. PROPOSAL OF NOISE BARRIERS between the junction with Kolodvorska Street at km 3.74 and the block access at km 3.85, there is a diverse area of We proposed noise barriers in the seven areas described four overexposed buildings (the house Kolodvorska Street above. Before making a proposal for noise barriers, we 8, two houses Prijateljev trg 7 and 8 and a 5-storey detailed all the input data for the calculations in the area residential block Prijateljev trg 6). The planned land use of of the proposed barriers, calculated the economic the area with houses is SS - residential, the same applies eligibility index for the different barrier heights and to the block. At km 3.775 there is a street, which decided on the appropriate barrier height based on the (according to the usual curb on the pavement along the values. In total, seven sets of barriers were proposed in main road) is no longer accessed from the main road but seven areas, respectively. One barrier is 3 m above the can certainly be accessed from the other side (from the road, two barriers are 2.5 m high and the remaining south). There is currently a pedestrian access at ca. km barriers are 2.0 m high. The barriers have a total length of 3.81. The terrain, including the road, is predominantly flat. 1303 m and a surface area of 2928.5 m2. The proposal of We propose the area for a noise barrier, although due to noise barriers is shown in table 1 and graphically on its central location in the settlement, the possible need for figures 1 to 6. The table gives information on the barrier pedestrian passages and the very different heights of the number (equal to the number of the corresponding area), buildings, we were somewhat hesitant to make this its length and height, its position in relation to the road decision. (stationing from - to, side) and the need for absorption Area 7 (see figure 6): In Goriča vas, there are several characteristics on the inside (towards the road) and overexposed residential buildings to the left between km outside of the barrier. 5.15 and 5.35. At km 5.22 there is a single junction to the left, otherwise there are no individual entrances from the Noise L H From To Side Abs. main road. From the junction, the road continues on its barrier ** In/Out (PHO) left side over a bridge (river Bistrica). Buildings are at or (m) (m) (km) (km) slightly below road level. The buildings are zoned SK - rural PHO 1 116 2,5 12,270 12,383 right +/-* settlement area, but in the part under consideration the PHO 2 258 2,0 0,637 0,891 left -*/- buildings are mainly residential. Unfortunately, the PHO 3 213 3,0 1,572 1,791 right +/- junction will require an interruption of the barrier, but in PHO 4*** 295 2,0 1,245 1,537 left +/- spite of this only interruption, we have proposed a noise PHO 5 79 2,0 3,664 3,737 left +/+ barrier, although the noise character (short higher levels PHO 6 103 2,5 3,743 3,842 right +/- at nearby houses when a vehicle crosses the interruption) PHO 7/1 68 2,0 5,155 5,220 left +/- might not be ideal. PHO 7/2 171 2,0 5,228 5,395 left +/- * Absorption is not obligatory, but is recommended, at least on some The study describes in the same detail the areas suitable parts of the barrier. for noise barriers and the areas not suitable for noise ** All heights are relative to the national road, except for PHO 2 up to approx. km 0.7 (there above the ground - the upper edge of the cutting) barriers. In this article, we have described the areas and PHO 7/1 and PHO 7/2 near the junction (there above the local road). suitable for noise barriers only. The reasons, why some *** The barrier PHO 4 can only be implemented at the same time as or after the reconstruction of the road, which removes the connections of areas were considered not suitable for noise barriers, are the existing local roads to the main road. summarized here: too few overexposed buildings, buildings high above the road, extraordinary views (to the Table 1. Proposed noise barriers mansion, the old town, the river), houses directly on/near the road, houses with individual driveways directly from Each of the barriers is described in detail in the study the road, most buildings not sensitive to noise, tall blocks (skipped in this article), in particular in terms of its route, with asphalted parking lots in the surroundings, scattered the start and end points, the possibilities of breaks for housing, old cemetery with stone wall under monument (pedestrian) crossings, and other features that might be protection proposing (the settlement has a new relevant for the design. The acoustical effect of noise cemetery), barriers are not possible for reasons of traffic barriers (Lnight with minus without barrier) is shown in the safety (in a curve, at an intersection). lower halves of figures 1 to 6. Žiger: Proposing noise barriers along existing national roads through settlements – the case of Novo mesto and its surroundings 254 AAAA – 2023 – IZOLA - Conference Proceedings 12+40 < -15 dBA 0 > -15 dBA 0 > -10 dBA 65 10180246 > -7 dBA 101 12 > -4 dBA 80 + 3 35 1 2 0 > -1 dBA 7 7 10180259 10179137 47 1 1 4 > 1 dBA 0 O 10179156 0 > 4 dBA PH 0 0 > 7 dBA 7 1 10179195 10180279 2 10179196 > 10 dBA +300 > 15 dBA 10180283 10179253 10179246 10180284 0 5 7 10180304 10179298 10180305 10180322 P 10180336 HO 2 0 80 10180354 10180363 10180374 < -15 dBA > -15 dBA > -10 dBA 0 5 8 10180390 > -7 dBA > -4 dBA > -1 dBA > 1 dBA > 4 dBA 10180435 10180431 > 7 dBA > 10 dBA 0 0 9 > 15 dBA Fig.1. Area 1 and barrier PHO 1 (above) and effect of Fig.2. Area 2 and barrier PHO 2 (above) and effect of barrier PHO 1 at h = 2 m (below), M 1:2500 barrier PHO 2 at h = 2 m (below), M 1:2500 Žiger: Proposing noise barriers along existing national roads through settlements – the case of Novo mesto and its surroundings 255 AAAA – 2023 – IZOLA - Conference Proceedings 22443537 < -15 dBA > -15 dBA 1 > -10 dBA +5 32216390 50 1 > -7 dBA +250 > -4 dBA 22413618 > -1 dBA > 1 dBA > 4 dBA 26696602 zacetek PHO 3 22413617 31538575 > 7 dBA 1+ 1+ 3 6 0 00 > 10 dBA 0 > 15 dBA 31538543 PH 31533056 O 1 4 +35 0 31535258 0 5 6 + 1 22413615 31535250 1+4 00 31535205 22413608 31 1 5 0 3 1 5 8 1 1 6 5 3 8 0 0 7 + 1 3 5 15 22413607 37986 PHO 1+ < -15 dBA 450 22413609 3 > -15 dBA > -10 dBA 31513547 26696520 > -7 dBA 0 5 7 + 1 > -4 dBA 31488736 > -1 dBA 22413591 1 > 1 dBA +50 0 > 4 dBA > 7 dBA > 10 dBA 0 0 > 15 dBA 8 + 1 1+5 50 Fig.3. Area 3 and barrier PHO 3 (above) and effect of Fig.4. Area 4 and barrier PHO 4 (above) and effect of barrier PHO 3 at h = 2 m (below), M 1:2500 barrier PHO 4 at h = 2 m (below), M 1:2500 approx. Žiger: Proposing noise barriers along existing national roads through settlements – the case of Novo mesto and its surroundings 256 AAAA – 2023 – IZOLA - Conference Proceedings 22411004 5+1 50 22411001 P 22411000 HO 5+ 22412665 200 7/1 22410999 3 P + HO 6 5 50 22412370 3 22410998 3 + + 7 7 5 0 0 0 5+2 3 50 +80 22411006 0 26700877 22411069 072 22412584 3+ P 8 H 5 O 0 6 22411 418 22412585 PHO 26701 7 631 0 0 3 + 5 22412583 /2 31320 22412586 074 31308136 22411070 22411 < -15 dBA < -15 dBA > -15 dBA > -15 dBA 22412589 > -10 dBA > -10 dBA 22411073 > -7 dBA 5+ > -7 dBA 350 > -4 dBA > -4 dBA > -1 dBA > -1 dBA > 1 dBA > 1 dBA > 4 dBA > 4 dBA > 7 dBA > 7 dBA > 10 dBA > 10 dBA + 5 0 0 4 > 15 dBA > 15 dBA Fig.5. Areas 5 & 6 and barriers PHO 5 & 6 (above) and Fig.6. Area 7 and barrier PHO 7 (above) and effect of their effect at h = 2 m (below), M 1:2500 barrier PHO 7 at h = 2 m (below), M 1:2500 Žiger: Proposing noise barriers along existing national roads through settlements – the case of Novo mesto and its surroundings 257 AAAA – 2023 – IZOLA - Conference Proceedings It was not possible and reasonable to protect all buildings They have a total length of 1303 m and a surface area of in their entirety by noise barriers, so we also proposed 2,928.5 m2. The proposed noise barriers have a good or passive protection in the noise barrier areas. Passive very good economic eligibility index and are going to protection was proposed in areas 1, 2, 3 and 6 for 15 reduce noise significantly. Passive protection was residential buildings, or rather for their upper floors. No proposed for the upper floors of 15 buildings. passive protection was required in areas 4, 5 and 7. In this article, we have shown that there are viable options We have not proposed quieter asphalt as it has a for noise barriers along existing main and regional roads negligible acoustic effect compared to barriers at through settlements. Although noise barriers proposals residential traffic speeds (50 km/h). We have also not always are somewhat subjective, they should proposed lower speed limits, as these are not effective nevertheless be based on common baselines, e.g. such as from a noise point of view at 50 km/h. However, we have we have outlined in this article. recommended that quieter asphalts should be used when resurfacing is necessary. As part of the study, noise measurements were also 7. REFERENCES carried out in some areas of the proposed noise barriers, in order to be able to assess the effect of each noise [1.] Žiger, M. (NLZOH) Noise study for the identification barrier after its implementation and to compare it with of areas and dimensions of necessary noise barriers along important roads managed by the the expected results provided by this noise study. Infrastructure Directorate of the Republic of Slovenia in the area of Novo mesto and its surroundings, No. 2920-21/97833-22, 2022 (in 6. CONCLUSION Slovene). [2.] PNZ d.o.o., Epi Spektrum d.o.o. and A-projekt d.o.o. We presented a noise study to identify the areas and Noise monitoring for roads with more than 3 million extent of noise barriers along major roads managed by the vehicle movements per year, managed by the Infrastructure Directorate of the Republic of Infrastructure Directorate of the Republic of Slovenia in Slovenia, No. 17/650A, 2019 (in Slovene). the area of Novo mesto and its surroundings, which was [3.] Epi Spektrum d.o.o. and PNZ d.o.o. Expert basis for carried out at NLZOH in 2021 and 2022. the operational program for noise abatement along The total length of the national roads concerned was motorways in the Republic of Slovenia, No. 2018- around 15 km. As part of the study, road traffic noise 015/IMS, 2019 (in Slovene). emissions for 2045 were determined and an acoustic [4.] Data on roads (DRSI), municipality plans and model developed. Based on the baselines presented, we conditions (municipalities Ribnica and Novo mesto), general environmental and geographical data identified seven areas where noise barriers make sense. (internet), 2021 and 2022, and project We developed a proposal for noise barriers and associated documentation for reconstructions of individual road passive protection. In total, seven sets of barriers (one 3 sections, all used for noise modelling. m above the road, two 2.5 m high and the remaining ones 2.0 m high) were proposed in seven areas, respectively. Žiger: Proposing noise barriers along existing national roads through settlements – the case of Novo mesto and its surroundings 258 MEASUREMENT AND MODELLING UNCERTAINTY IN ACCREDITED ACOUSTIC PROCEDURES Antonio Petošić, Domagoj Stošić, Toni Marinković, Mia Suhanek University of Zagreb, Faculty of Electrical Engineering and Computing, Department of Electroacoustics, Unska 3, Zagreb Abstract: . The uncertainty in acoustic measurements and modelling is discussed in this article regarding the measurement method for determination of environmental noise parameters in accordance with ISO 1996-2:2017, the modelling methods in accordance with ISO 9613-2:1997, the NMPB-XPS 2007, and the EU CNOSSOS from the results of interlaboratory comparisons (ILC). Additionally, measurement uncertainties for airborne and impact building acoustic parameters measured in accordance with ISO 16283-1:2014, ISO 16283-2:2020, with determination of single number values in accordance with ISO 717-1:2020 and ISO 717-2:2020, from ILC are determined from each independent measurement and compared with tentative values from the ISO 12999-2020 standard and previous comparisons. The statistical analysis of the data shows that there is a significant difference in the proposed measurement uncertainty for equivalent sound pressure level at closer ranges. Modelling uncertainty from the findings is significantly less than the values provided by laboratories which use tentative value for umet=2 dBA which is the most dominant component in uncertainty budged. When discussing building acoustic parameters, there is no discernible difference between the ILC findings obtained with one independent measurement and those recommended in standard ISO 12999-1:2020 and previous comparisons. Keywords: measurement and modelling uncertainty in acoustics, environmental noise parameters, building acoustic parameters, experimental measurement uncertainty standard deviations in repeatability and reproducibility conditions; 1. INTRODUCTION For some other procedures, e.g., in building acoustics, there is a separate standard ISO 12999-1:2020 Finding measurement uncertainties in acoustic [2] which considers the measurement uncertainty for measurements and modelling can be a very tedious task airborne and impact and sound insulation parameters by due to large number of parameters which have an using values for standard deviations in repeatability and influence on measurement results. In environmental reproducibility conditions. measurement accredited procedure described in In some other standard for determination of the standards ISO 1996-2:2017 [1] there is a chapter about sound power such as ISO 8297:1997 [3] the measurement measurement uncertainty budget. The uncertainty for uncertainty is predefined depending on the size of the LA,eq and some other parameters can be calculated for measurement site (Table 1). The maximum value is up to each considered situation and different sources levels U=+1,5 dBA (single side interval 95% confidence level). under the influence of residual noise levels. The In the standards ISO 3744:2010 [4] and ISO measurement results are usually acquired in small time 3746:2010 [5] the measurement uncertainty is also intervals (few hundred samples in 10 minutes) if longer provided, and usually a maximum value is given (U=4 term measurements are done or few minutes intervals if dBA). short term measurements are done. Most of laboratories Measurement uncertainty for standard ISO 3382- do short term measurements at shorter distance where 1,2 [5,6] depends on the measurement results and meteorological conditions do not play a significant role on number of measurements (source and receiver positions). the measurement results. Petošić et al.: MEASUREMENT AND MODELLING UNCERTAINTY IN ACCREDITED ACOUSTIC PROCEDURES 259 AAAA – 2023 – IZOLA - Conference Proceedings There are some additional procedures to estimate measurement uncertainty u (repeating the measurements and calculating standard deviations in repeatability conditions) and finding expanded measurement uncertainty u. Due to simplicity, the majority laboratories used the maximum predefined values for uncertainties from standard ISO 19962-2:2017. Usually, they did not use the uncertainties in assessment of conformity with limit values. In this paper the main problems in measurement uncertainty determination for measurement procedures for environmental noise and building acoustics parameters is considered. Experimental measurement uncertainties obtained by different laboratories participating in several ILCs and overall experimental measurement uncertainty are compared. 1.1. Measurement uncertainty for environmental noise parameters by measurements and modelling In the first step usually the background noise level is measured. The measurements should be done in Fig.1. Procedure for determination of environmental most favourable meteorological conditions, if possible, noise parameters according to ISO 1996-2:2017 from the dominant source of residual noise (highway, (measuring initial state, modelling new sources state, railway, road far away or industrial site). This can be a checking the results) problem if the residual noise is measured in the state when it is not lowest during the day, evening, night period It is evident that the new edition from 2017 is so sometimes LA,90 is used as a parameter for estimating more complex than previous version ISO 1996-2:2007 the residual noise level instead LA,eq [7]. regarding the measurement uncertainty calculations. The measurement procedure for determination Each measurement situation and time of measurement of environmental noise parameters by measuring of can be subjected to different operating conditions from residual noise, modelling the levels of new source, and source and different meteorological conditions and checking modelling results by measurements of overall therefore, measurement uncertainty cannot be fully noise when new source should be installed near controlled by the operator. The standard covers all types residential objects is shown in Fig. 1. of sound sources (traffic, industry) and meteorological conditions (very favourable, favourable, neutral, and unfavourable). For some meteorological conditions (unfavourable) results should be omitted (problems with short term measurements on larger distance than eq. 11 in ISO 1996-2:2017) even this is very difficult to achieve during day period. The possible strategies for measurements are shown in Figs. 2 upper and down. Majority of laboratories have done short term measurements with tracking meteorological conditions for both sources due to restriction of time for participation in ILC (several hours for environmental noise measurements and few hours for building acoustics parameters). Petošić et al.: MEASUREMENT AND MODELLING UNCERTAINTY IN ACCREDITED ACOUSTIC PROCEDURES 260 AAAA – 2023 – IZOLA - Conference Proceedings ƍmet 0,0 2,0 (1,5) 1,00 2,00 ƍ loc 0,0 0,00 0,00 ƍ res 0,4 -0,01 -0,01 L A,eq,res 39,1 2 2 u=sqrt(u +u +..) 1 2 2,08 Expanded L , A,eq corrected 58,2 k=2 U=4,2 (3,2) *the largest component in uncertainty budget Table 1. Measurement uncertainty budget according to [1] In the ILC two different sources have been considered: stable industrial source with tonal component at 10 kHz and local road as a source of residual noise. The receivers’ height for stable source was from 1.2 metres to 1.5 metres and for a local road (total length 4 metres). The scheme of setup is shown in Fig. 3a). while Fig. 3b) shows the modelling site. Fig 2. Possible strategies for measurements in ILC (upper) with some restrictions in (down) due to available time. The process of measurement uncertainty estimation is much more complicated (influence of all parameters, equipment, repeatability conditions, different operating, and meteorological windows). The standard aims to cover all possible conditions however, it is a basis for developing more specific standards dedicated to the specific type of sound sources and goals. A more complex way to calculate measurement uncertainty would be dividing the sound source work in different operating and meteorological conditions. The uncertainty budget is shown in Table 1. and it is slightly different from previous standard edition. Standard Senistivity Uncertainty Quantity Estimate Uncertainty Coefficient Contribution L' A,eq 58,3 0,5 1,01 0,51 Fig 3. Measurement setup for ILC with noise sources and ƍ slm 0,0 0,00 measurement positions a) and modelling site b) ƍ sou 0,0 0,3 1,00 0,28 Petošić et al.: MEASUREMENT AND MODELLING UNCERTAINTY IN ACCREDITED ACOUSTIC PROCEDURES 261 AAAA – 2023 – IZOLA - Conference Proceedings Two measurement points for a stable sound source The results for measurement uncertainty from each (located in the house with one window-surface source laboratory and overall obtained in ILC have been opened) are considered: free field (on the border 25m compared. away) and in front of the facade (50m) and on the Modelling the environmental noise parameters facade/near facade with corrections for all measured around the surface (radiating window) and local road as a parameters (-3 dB). ‘’line source’’ is done by knowing the properties of terrain All laboratories had to measure environmental and objects. Modelling methods with all propagation parameters from the stable industrial source at two effects are: ISO 9613-2:1997, CNOSSOS –EU for industrial different positions (several short-term measurements i.e., source and EU-CNOSSOS and NMPB 2088 for local road. several intervals to “cover” a 15-minute interval) to satisfy The points for providing results are given by the organizer. (or not) the conditions from eq. 11 in ISO 1996-2:2017 The estimated sound source power (point source when meteorological windows have to be determined. and line source) together with sound pressure levels at Laboratories which were measuring meteorological imission points of interest are calculated by the conditions, provided the measured meteorological data participants while using measurements in two different during measurements with determined meteorological points. windows (M1,M2,M3,M4) connected with noise samples and were compared with organizer meteorological data. 2.3. Sound insulation setup For the local road the sound exposure level ( LAE) [7] is measured during propagation of each individual type of The measurement setup is shown in Fig. 4. with vehicle (passenger, medium-heavy, heavy…) on one source and receiving rooms for airborne and impact sound measurement position. If this was not possible, the insulation measurements according to ISO 16283-1:2016 laboratories measured the A-weighted equivalent sound [8] and ISO 16283-2:2020 [9] with single number values pressure level in some period with given number of determination together with spectral adaptation terms vehicles in order to find sound exposure level from described in ISO 717-1:2020 [10] and ISO 717-2:2020[11]. individual pass-bys of different types of vehicles. The The laboratories determined measurement uncertainties typical duration of measurement was 1 hour. according to ISO 12999-1:2020 [12]. The levels of background noise from highway (500 m) were different at different measurements intervals due to increased traffic flow or favourable meteorological conditions from line source because highway as a source was main source of background noise. The critical questions arise: a) Is the measurement uncertainty overestimated when umet=2 dBA (for LA,eq) and should we consider uncertainty when we are doing assessment? b) How to measure meteorological conditions during environmental noise parameters (resolution in time, which parameters to measure and use meterological windows or more detailed considerations at all wind directions)? Fig 4. Measurement site for sound insulation parameters c) Is it better to use results of interlaboratory comparisons results for uncertainty or to use The same receiving room was used for airborne GUM when expanded measurement uncertainty and impact sound insulation measurements due to time is calculated from each individual measurement restriction. Only one independent measurement is done results according Table 1.? with measuring all parameters which are used in sound insulation parameters calculations. Petošić et al.: MEASUREMENT AND MODELLING UNCERTAINTY IN ACCREDITED ACOUSTIC PROCEDURES 262 AAAA – 2023 – IZOLA - Conference Proceedings A new method for estimating measurement uncertainty considers standard deviations of measured SPL levels in the source and receiving room, reverberation times and influence of background noise and measurement equipment uncertainty. In addition, the results for R’, L’n, in a frequency range from 50Hz to 5kHz with all spectral adaptation terms and determined measurement uncertainties for each weighted parameter. 2. MEASUREMENT RESULTS The measurement uncertainties for main measured parameters are shown (for each part of ILC-2019) in subsections. 2.1 Environmental noise parameters measurement results The residual noise results are shown in Fig. 5. with experimental measurement uncertainty ures and usou. Fig. 5. Measurement results for LA,eq with ures for residual noise and rating levels at two different positions (free field and close to the façade for stable industrial source). The applied measurement uncertainties from each laboratory were slightly higher than u=2 dBA due to umet and overall experimental measurement uncertainty for each position and parameter which is shown in Table 2. AVG Number of u Parameter (dB) STDEV (dB) participants (dBA) L'A,eq-Residual 43,6 noise -S1P1 2,1 41 0,3 L'A,eq-Residual 44,4 noise -S1P2 2,3 41 0,4 LRA,eq- Source 56,1 noise- S1P1 0,9 41 0,1 LRA,eq-Source 49,4 noise S1P2 1,1 41 0,2 Table 2. Reported measurement uncertainty for measured parameters at two different locations (overall results) In addition, for local road measurement results are shown per parameter with experimental measurement uncertainty in Table 3. and logging procedure is shown in Fig. 6. The logging in 1 second interval has been done and furthermore, the vehicle Petošić et al.: MEASUREMENT AND MODELLING UNCERTAINTY IN ACCREDITED ACOUSTIC PROCEDURES 263 AAAA – 2023 – IZOLA - Conference Proceedings categories have been detected and LAE parameter has power of the source) and then that data was used to been found for each vehicle category. calculate the noise levels at other proposed locations. 2.2.1. Modelling results for the stable source The results for stable source sound power with modelling uncertainty are shown in Table 4. AVG -ISO u(dB)- AVG-EU- 9613- ISO u (dB)-EU CNOSSOS 2:1997 9613- CNOSSOS (dB) (dB) 2:1997 S1-power 88,2 1,2 90,6 1,3 Fig 6. Example of logging the LAeq,1s from each vehicle Ld-1 45,4 0,6 46,4 0,4 category L d-2 45,3 0,7 44,1 0,6 Ld-3 51,5 1,5 51,9 1,0 AVG STDEV Number of Ld-4 48,6 0,6 48,2 0,6 Parameter (dB) (dB) participants u (dBA) LRA,eq 45,6 2,3 20 0,5 Table 4. Overall results and experimental L' uncertainties for stable industrial source obtained by two A,95 38,9 2,8 19 0,6 L modelling methods AE -L 50,7 3,1 15 0,8 L AE -M 55,4 2,3 16 0,6 L It is evident that approximately the same AE -H 60,8 3,2 16 0,8 L modelling uncertainties are obtained for two different A,max -L 48,7 3,8 16 1,0 L methods ISO 9613-2:1997 [13], EU-CNOSSOS [14] and A,max -M 50,5 3,6 16 0,9 NMPB 2008 [15] and in addition, when compared with LA,max -H 558 3,9 16 1,0 reported modelling uncertainties from laboratories they Table 3. Experimental measurement uncertainty for local are much lower ( u=3 dBA for sound power and road parameters at 100m distance in free field approximately u=4 dBA for Lday). measurement location 2.2.2 Local road source The meteorological windows were calculated from each laboratory and compared to the organizer and In this part of ILC, two different approaches have during the day period only M1 and M2 meteorological been tested. In the first approach, laboratories used windows are obtained. measurement data at S2P3 position (source 2 position 3 in free field) to calculate noise levels at other positions with There were a lot of problems with correction of the number of vehicles during measurement from each the results due to positions (residual noise and source lab (different time, different traffic density, different levels, corrections due to background noise, adding tonal meteorological conditions) . In the second approach, only penalty because a tone was added at 10 kHz). the model has been calibrated with the measurement results and the organizer provided the referent number of 2.2 Modelling noise parameters results vehicles (day, evening night period) to calculate sound descriptors. The laboratories used measurement results at The results for number of vehicles during closer position to calibrate the model (determine sound measurements, sound power is and descriptors Ld for two positions during the day period are shown and Table 5. Petošić et al.: MEASUREMENT AND MODELLING UNCERTAINTY IN ACCREDITED ACOUSTIC PROCEDURES 264 AAAA – 2023 – IZOLA - Conference Proceedings road source sound power ( u) from 2 dBA up to 2,5 dBA, and sound descriptors from 2 dBA to 3,5 dBA. AVG u(dB)- AVG All laboratories measured noise levels in neutral u(dB)- (dB)- NMPB- (dB)- on unfavourable conditions during day period and have CNOSSOS NMPB XPS CNOSSOS not corrected the modelling results according to favourable or very favourable conditions. S2-power 79,9 0,8 77,7 0,6 The PDF distribution of the results is shown in Fig. Ld-1 46,8 0,9 44,3 0,7 7. and it can be noted that the distribution is almost Gaussian for all measured parameters. Ld-2 43,4 1,0 41,5 0,6 Ld-3 40,8 1,8 38,1 1,6 Ld-4 41,6 1,5 40,0 1,0 Table 5. Modelling uncertainty for different methods of calculation (NMPB-XPS and CNOSSOS EU-8). The same procedure is repeated for defined number of vehicles during different periods and the results for modelling with different methods are shown in Table 6. AVG- u(dB)- AVG- u(dB)- NMPB NMPB CNOSSOS CNOSSOS S2- power 73,9 0,6 79,4 0,6 Ld-1 49,0 0,5 46,1 0,6 Ld-2 45,5 0,6 43,2 0,5 Ld-3 43,3 1,2 40,8 1,2 Ld-4 44,6 0,8 42,0 1,1 Table 6. Modelling results and uncertainties with given number of vehicles during day (Nlight=600 /hour , Nmedium= 48 /hour, Nheavy=12 /hour). It can be observed that with the same data the sound descriptors obtained by CNOSSOS method are lower than compared to the NMPB-XPS method. The modelling uncertainty for all results is significantly lower than reported from the participants (for stable and local Petošić et al.: MEASUREMENT AND MODELLING UNCERTAINTY IN ACCREDITED ACOUSTIC PROCEDURES 265 AAAA – 2023 – IZOLA - Conference Proceedings It is evident that the reported measurement uncertainties (U=-0,9 dBA and U=+1,5 dB) are much higher than in cases of considering all measurement results (i.e., without outlier) assuming no correlation between the results (uairborne=0,13 dB and uimpact=0,14 dB). 3. CONCLUSION Fig 7. The PDF distribution of measurement and modelling results without outliers for stable and local It can be concluded that the overall experimental road source uncertainty for measurement and modelling environmental noise parameters is significantly lower It is possible that the difference in shape has than those provided by laboratories (experimental occurred due to the directivity of the source at closer measurement uncertainty u is slightly higher than 2 dBA position to the industrial source (main lobe width). according to ISO 1996-2:207 standard). It is very difficult to estimate if the residual noise will be increased more 2.3. Sound insulation parameters measurement than +1 dBA if the residual noise levels are close to the uncertainties overall levels (source+background) because according to new standard the conditions from standard ISO 1996- The obtained single number values with reported 2:2017 are not satisfied in that case (3dBA difference). measurement uncertainties U (k=1, 95% confidence level) The reported modelling uncertainty is much are shown in Fig. 8. higher compared to the case of averaging all results and finding standard deviations and experimental measurement uncertainty from all results. Regarding sound insulation, the single number uncertainty is at the highest point when values from ISO 12999-1:2020 standard are used to find upper and down curve for sound insulation compared to the situation when individual uncertainties are found from each independent measurement results for all parameters in one-third octave bands. The main problem was to check and correct reported results from laboratories in order to obtain a real comparison because laboratories provided results with several mistakes (e.g., measuring point not in the middle of the main lobe, measuring samples from source far away in time from residual noise samples, using A-spectrum in building acoustics measurements, calibration of model not in meteorological conditions which appeared during measurements etc.) Finally, the criterium for z-score is obtained assuming that there is no correlation between the results (ISO 5725-2:2019) [13] while the results between Fig 8. The obtained results for sound insulation laboratories are obviously correlated. parameters with reported measurement uncertainties (standard ISO 12999-1:2020) 4. REFERENCES [1.] ISO 1996-2:2017, Acoustics - Description, measurement and assessment of environmental Petošić et al.: MEASUREMENT AND MODELLING UNCERTAINTY IN ACCREDITED ACOUSTIC PROCEDURES 266 AAAA – 2023 – IZOLA - Conference Proceedings noise -- Part 2: Determination of sound pressure [10.] ISO 717-1:2020 Acoustics, Rating of sound levels, Geneva (Switzerland): International insulation in buildings and of building elements – Organization for Standardization. Part 1: Airborne sound insulation, Geneva [2.] ISO 12999-1:2020 Determination and handling of (Switzerland): International Organization for uncertainties in building acoustics, Part 1: Sound Standardization. insulation, Geneva (Switzerland): International [11.] ISO 717-2:2020 Acoustics, Rating of sound Organization for Standardization. insulation in buildings and of building elements – [3.] ISO 8297:1994 :Acoustics — Determination of sound Part 2: Impact sound insulation, Geneva power levels of multisource industrial plants for (Switzerland): International Organization for evaluation of sound pressure levels in the Standardization. environment — Engineering method, Geneva [12.] ISO 12999-1:2020, Determination and handling (Switzerland): International Organization for of uncertainties in building acoustics. Part 1: Sound Standardization. insulation, International Organization for [4.] ISO 3744:2010, Acoustics — Determination of sound Standardization. power levels and sound energy levels of noise sources [13.] ISO 9613-2:1996 Acoustics — Attenuation of using sound pressure — Engineering methods for an sound during propagation outdoors — Part 2: essentially free field over a reflecting plane, Geneva General method of calculation, International (Switzerland): International Organization for Organization for Standardization. Standardization. [14.] Joint Research Centre, Institute for Health and [5.] ISO 3382-1:2009, Acoustics — Measurement of room Consumer Protection, Anfosso-Lédée, F., Paviotti, M., acoustic parameters — Part 1: Performance spaces, Kephalopoulos, S., Common noise assessment Geneva (Switzerland): International Organization for methods in Europe (CNOSSOS-EU) : to be used by the Standardization; 2010. EU Member States for strategic noise mapping [6.] ISO 3382-2:2008-Acoustics — Measurement of room following adoption as specified in the Environmental acoustic parameters — Part 2: Reverberation time in Noise Directive 2002/49/EC, Publications Office, ordinary rooms, Geneva (Switzerland): International 2012, https://data.europa.eu/doi/10.2788/32029 Organization for Standardization. [15.] Dutilleux, G.; Defrance, J.; Ecotière, D.; Gauvreau, [7.] ISO 1996-1:2017, Acoustics - Description, B.; Bérengier, M.; Besnard, F.; Duc, E.L. NMPB-routes- measurement and assessment of environmental 2008: The revision of thefrench method for road noise --Part 1: Basic quantities and assessment traffic noise prediction, Acta Acust. United Acust. procedures, Geneva (Switzerland): International 2010, 96, 452–462. Organization for Standardization. [16.] ISO 5725-2:2019 Accuracy (trueness and [8.] ISO 16283-1:2016: Acoustics -- Field measurement of precision) of measurement methods and results – sound insulation in buildings and of building elements Part 2: Basic method for the determination of -- Part 1: Airborne sound insulation, Geneva repeatability and reproducibility of a standard (Switzerland): International Organization for measurement method, International Organization Standardization. for Standardization. [9.] ISO 16283-2:2020: Acoustics -- Field measurement of sound insulation in buildings and of building elements -- Part 2: Impact sound insulation, Geneva (Switzerland): International Organization for Standardization. Petošić et al.: MEASUREMENT AND MODELLING UNCERTAINTY IN ACCREDITED ACOUSTIC PROCEDURES 267 DEVELOPMENT OF A SPECIAL STANDARD FOR OUTDOOR MUSIC EVENTS IN SLOVENIA BASED ON MEASUREMENTS AND CALCULATIONS Luka Čurović, Jure Murovec, Nejc Cerkovnik, Anže Železnik, Jurij Prezelj University of Ljubljana, Faculty of Mechanical Engineering Abstract: In Slovenia, the standard ISO 1996-2 is used to conduct the environmental noise measurements. When planning and performing measurements, a large number of variables must be taken into account, including the sound-emitting characteristics of the noise source, meteorological conditions, and the specifics of local geography and noise-sensitive areas. In general, the standard attempts to cover all types of sources, so it is best used as a basis for developing more specific standards for particular sources such as modern loudspeakers and outdoor sound systems. The objective of this paper is to present the acoustic characteristics of sound systems used at outdoor music events and to identify the main influencing variables that should be considered in the design and evaluation of environmental noise based on measurements and calculations. Such a study could be a first step in developing specific practice guidelines and standards for outdoor music events. Keywords: Environmental noise, outdoor events, loudspeakers, standard 1. INTRODUCTION premises, as well as the method of noise prediction and the monitoring procedure. The World Health Organisation (WHO) and the European Environment Agency (EEA) consider noise pollution, 2.1. Prediction using modelling including recreational noise, to be a major environmental problem [1]. An important source of recreational noise is Noise pollution or environmental noise pollution is outdoor entertainment events using modern sound calculated according to the simplified version of standard systems. Outdoor events are increasingly taking place in 9613-2 [4], taking into account only the geometric populated areas [2], and the number of complaints from divergence (A) using the following equation: the affected public has increased. The regulations in force in Slovenia on noise from musical events [3] and research 𝐿𝑓,𝑖 = 𝐿𝑊 + 𝐷𝑐 − 𝐴; 𝐴 = 20 × log(𝑑⁄𝑑0) + 11 (1) in this field [1] have not been able to successfully address all the noise problems associated with modern sound where 𝐿𝑓,𝑖 is the sound pressure level at receiver location reinforcement systems. due to i-th loudspeaker at frequency 𝑓, 𝐿𝑊 is sound power This paper discusses the existing regulations on level of the source, 𝐷 environmental noise which is emitted from events with 𝑐 is the directivity factor, 𝑑 is a distance between source and receiver and 𝑑 loudspeakers and presents specific problems related to 0 is a reference distance equal to 1 m. The total sound pressure modern music events. It highlights the main issues that level at receiver location is calculated using energetical need to be considered when evaluating the summation. environmental noise of music events based on measurements and calculations. 𝑁 (2) 𝐿 ) 2. CURRENT SITUATION IN SLOVENIA 𝑓 = 10 × log (∑ 100.1𝐿𝑓,𝑖 𝑖=1 In Slovenia, environmental noise caused by outdoor The sound power level of a loudspeaker should be events is currently regulated by legislation that sets the determined by measuring the sound pressure level at a permissible limits in front of buildings with residential distance of 1 m in front of the loudspeaker (in the axis of 268 AAAA – 2023 – IZOLA - Conference Proceedings the loudspeaker). The Slovenian regulation also on the spot or monitored by sound engineers, listeners, introduces the term electrical working power, which is affected neighbours, or local authorities. Labelling sound expressed as a percentage of the maximum electrical reinforcement systems by sound power level is also in power of the loudspeaker. direct conflict with the regulation. The noise study should contain a description and determination of the permissible electrical working 2.2. Directivity power, and the permissible electrical power should also be visible at the event. Legislation sets the requirements for directivity of The emission data of a loudspeaker can be expressed as loudspeakers. Currently, our legislation is based on a sensitivity (E) in dB/W/m, which corresponds to the sound fundamental theoretical analysis carried out by Lord pressure level in dB at a distance of 1 m from the Rayleigh in the nineteenth century. The directivity of loudspeaker and an input power of 1 W. The actual sound loudspeakers is based on the mathematical theory about pressure level is determined taking into account the the sound field in the far field generated by a circular power supplied by the amplifier P, according to: piston (loudspeaker diaphragm) in an infinite baffle. According to the theory, the expression for far-field sound 𝐿𝑝 = 𝐸 + 10 × log(𝑃⁄𝑃0) (3) pressure at a given distance 𝑟, polar angle 𝜃 (angle between the direction of the receiver point and the main Where 𝑃 axis of the loudspeaker), frequency 𝑓, and piston diameter 0 is the reference power, which is equal to 1 W. For example, the sound pressure level at 1 m in front of a 2𝑎. The directivity factor 𝐷𝑐 can be expressed as follows: loudspeaker with a sensitivity of 105 dB/W/m at an input power of 30 W is 105 dB + 10 log(30W/1W) = 120 dB. This 2 × 𝐽 (4) 𝐷 1(𝑘 × 𝑎 × sin 𝜃)) = 20 equation is usually used to solve an inverse problem. 𝑐 = 20 × log ( 𝑘 × 𝑎 × sin 𝜃 Based on the location of the loudspeaker, the location of 2 × 𝐽1(𝑧) the noise-sensitive receiver, and the allowable sound × log ( ) 𝑧 pressure level, the sound pressure level at 1 m is determined using geometric divergence. The sound where 𝐽 pressure level and sensitivity data are then used to 1 is first order Bessel function and the quotient 2𝐽 determine the allowable electrical power for the selected 1(𝑧)/(𝑧) shows the directivity function of the circular piston as a function of the wave number 𝑘 at 800 Hz, loudspeaker. loudspeaker radius 𝑎 and an 𝜃 angle between the For example, if the limit value at 20 m from the speaker direction of the receiver point and the main axis of the with sensitivity of 105 dB/W/m is set at 70 dBA. The loudspeaker. allowed electrical power is calculated by first determining the sound pressure level at 1 m: 2.3. Emission spectra 𝐿𝑝(𝑑 = 1𝑚) = 70𝑑𝐵 + 20 × log(20 𝑚⁄1 𝑚) = 96 𝑑𝐵𝐴 According to the current regulation, the directivity and sound propagation are calculated only at 800 Hz or at a 𝑃 = 𝑃0100.1(𝐿𝑝−𝐸) = 0.12 𝑊 frequency where the peak of the sound power level spectrum of the loudspeaker is specified by the The power supplied to such a loudspeaker would have to manufacturer using equation 4. In this way, the frequency be less than 0.12 W, which is of course irrelevant for characteristics of the sound produced by modern sound practical applications where amplifier power is much reinforcement systems are not taken into account in the greater (over 500 W). Such a procedure is not common in current regulation. other countries. For this reason, in practice, loudspeaker systems are often characterized by their maximum sound power levels. The 2.4. Measurements sound power levels are determined from environmental noise calculations according to 9613-2. Often the Compliance is rarely verified by measurements. In fact, calculations are performed in a single 1/3-octave measurements are rarely commissioned by local frequency band (usually 500 Hz) with directional authorities. Rather, measurements are commissioned by characteristics that are not usually specified. affected neighbours and performed by acoustic However, even if characterization of sound reinforcement laboratories with or without the knowledge of the music systems by sound power level can be considered a promoters. The ordinance does not explicitly specify the standard practise in acoustics, its use for monitoring is measurement standard or who is authorized to perform very impractical because it cannot be easily determined the measurements. It specifies the measurement location, 1st Čurović et al.: Development of a special standard for outdoor music events in Slovenia based on measurements and calculations 269 AAAA – 2023 – IZOLA - Conference Proceedings which must be chosen in front of affected buildings Hz was used as the excitation signal. The distance according to the standard ISO 1996-2 [5]. between the microphone and the loudspeaker was kept Noise limits, referred to as critical loads, are set according constant at 1 m. The loudspeaker was rotated from 0 to to the area where the measurement point is located and 360 degrees with a resolution of 22.5 degrees. The vary for daytime, evening, and nighttime hours, as well as measurement results were compared with theoretical for the duration of the event. calculations at 800 Hz using Equation 4. The results of the measurements and theoretical calculations are shown in If the event is shorter than 8 hours, the integration time is Fig. 1. The directivity factor is shown in a polar diagram equal to 8 hours. For longer events, the time duration is where the calculated and measured values were equal to the duration of the event. However, the normalised to the major axis of the loudspeaker, which is integration time is often chosen by the acoustic experts determined by the polar angle of 0 degrees. performing the measurements. Due to the long integration time and the measurement positions, which may be located at a great distance from the event, post-processing and quality assurance of the acquired acoustic data is a time-consuming process. In addition, the measurement report must be produced within 30 days of the event, which is often seen as an intentional delay by stakeholders that include the event organisers, neighbours, and local authorities. 3. STATE OF THE ART The current assessment of environmental sound levels at outdoor events where amplified music is played does not address all relevant issues. The specific issues and state of the art related to modern music events were explored through a literature search, with a technical document published by AES as the primary source of information. 3.1. Directivity Most modern‐ large‐scale sound reinforcement systems Fig. 1: modelled and measured directivity of a consist of flown line arrays [99], in which identical arrays loudspeaker at 800 Hz. are placed symmetrically on the sides of a stage, and a subwoofer system designed to reproduce the lower end The theoretical directivity cannot model the measured of the frequency spectrum (below 100 Hz). The subwoofer directivity and predicts smaller directivity pattern. The system is usually used as a horizontal array. Sound theoretical directivity predicts 𝐷𝑐= -2 dB in the 900 engineers spend a great deal of time configuring the directions while the measurements show that the sound reinforcement system to achieve uniform sound directivity 𝐷𝑐 = -6 dB. pressure distribution and frequency response throughout the audience area) This is achieved by placing the 3.2. Influence of low frequency spectra loudspeakers above the audience and optimizing their placement according to their characteristics such as Several studies [284] show that annoyance is significantly directivity, filtering, time delay, etc. Optimization is greater when a sound contains strong low‐frequency performed using various tools such as EASE, EASE Focus, components. Live music can be characterized by the low‐ ArrayCalc, Sound Vision, Mapp Online, etc., based on the frequency content. Furthermore, when low‐frequency coherent addition of sound sources. sounds are cantered in the range that causes chest resonance (around 50 Hz), this has health implications To investigate directivity, we performed simple directivity (around 50 Hz); this can cause whole‐body vibrations that measurements. The measurements were performed in an can lead to annoyance and/or discomfort. In such cases, a anechoic chamber at the Faculty of Mechanical reduction of the sound pressure level is required [76]. Engineering in Ljubljana. A Yamaha HS 5 loudspeaker was used. An E-sweep signal in the range from 20 Hz to 20000 1st Čurović et al.: Development of a special standard for outdoor music events in Slovenia based on measurements and calculations 270 AAAA – 2023 – IZOLA - Conference Proceedings The frequency content of various music genres was Most regulations at the national, regional, and local levels studied. A Yamaha HS 5 loudspeaker was used, and the are based on LAeq with specified integration times. music was recorded in an anechoic chamber 1 m from the Regulations may also limit the duration of events, the loudspeaker. The frequency spectra (third octave, number of events, and the time periods between normalized to 100 dB) between 31.5 Hz and 16000 Hz for successive events. In general, limits for regular music different music clips are shown in Figure 2. Electronic events in residential areas range from 50 to 75 dBA. In music and heavy metal music are characterized by low some cases, limits are set in relation to the background frequency humps. Classical music and brass music are noise level (usually between 3 and 15 dB above the characterized by frequency spectra that follow the shape residual level). Noise limits for larger events can be 85 dBA of A weighting filter. or more in residential areas, but are generally limited to the hours up to 11 p.m. Some countries (Denmark, Netherlands, United Kingdom, France, Ireland, Slovakia) set additional limits for LCeq or Leq in the octave range from 63 Hz to 125 Hz to account for low-frequency noise. In general, noise in the lower frequency bands is limited to below 80 dB. 3.6. Measurements: Noise monitoring is considered one of the most important tools for controlling sound emissions to the environment Fig. 2: frequency content of different music genres and has become ubiquitous at large outdoor (and indoor) entertainment events. One of the key challenges in noise 3.3. Intermittent noise measurements is that there is rarely only one noise source in an environment. Outdoor measurements taken in front Moreover, noise is more annoying when it is intermittent of the façade of a building and at a height of 1.2 to 1.5 m rather than continuous. The noise characteristics of above the ground, according to ISO 1996-2, are usually musical events are obviously characterized by the contaminated by residual noise. duration of the songs, the phase of the event (e.g., the To ensure accuracy and repeatability, measurements beginning, middle, and end of a musical event), and the should be made under stable meteorological conditions, time interval between successive events. which is not always possible. The measurement campaign should include monitoring at the front-of-house mixing 3.4. Integration time position (FOH) as well as at selected locations in the residential area, and the integration times should be The choice of integration time depends on the dynamics synchronized [21]. of the observed acoustic signal. It also has an important Ideally, noise measurements should be continuous and influence on the limits that can be set by regulatory conducted some time before, during, and after the event. authorities. Integration times for musical events range The results should be available to sound engineers, event from 1 minute to 240 minutes in different countries. organizers, and local authorities in real time. Limits based on short averaging times (1 – 5 minutes) Measurements should also be used in the preparation could cause engineers to be overly sensitive to the phase to calibrate prediction models and optimize sound measurement data, while long integration times of over system setup based on measured sound pressure levels. 60 minutes do not follow the dynamics of the event and Measured data should include LAeq, Leq, and LCeq levels, allow for greater variability between different parts of the as well as frequency spectra. Sound recordings could also event. Long integration constants are impractical and do be helpful in identifying unwanted noise. not allow real-time adjustment of the sound reinforcement system. A compromise must be found. In 3.7. Environmental noise prediction most cases in Europe, a 15‐minute integration time is used in practice along with real‐time monitoring. A 15‐minute Events using outdoor loudspeakers may be held if integration time corresponds to about three songs. immission levels at selected receiver locations are below critical levels. Compliance can be demonstrated through a 3.5. Regulation noise study that predicts environmental noise levels using a selected calculation model. 1st Čurović et al.: Development of a special standard for outdoor music events in Slovenia based on measurements and calculations 271 AAAA – 2023 – IZOLA - Conference Proceedings The most commonly used calculation model for 5) meteo data as percent of favourable conditions, environmental noise in Europe is described in Annex II of 6) number of reflections, the Environmental Noise Directive. 7) optimization parameters (such as search distance or fetching radius, dynamic error Noise modelling is used to predict the propagation of margin, terrain interpolation, etc.) sound from the source to the receiver. The sound pressure level at the receiver is equal to the The calculation is standardized, but the results depend difference between the sound power level of the source mainly on the selected geographical, acoustic and and the summed attenuation factors, similar to equation calculation parameters. The model is based on the specific 1, where the attenuation is the attenuation due to parameters used in the calculation. This means that if the geometric divergence, atmospheric absorption, the characteristics of sound sources change, the results may ground, diffraction under favourable or homogeneous not be reliable. conditions. Some examples of the influence of the choice of input The source is characterized by the source type and its parameters are shown in Figures 3-5. The choice of acoustic properties. The model includes a topographic emission spectra with the same dBA level at the reference model with acoustically relevant objects such as buildings, point (LAeq = 80 dBA) on dBA and dBC levels is shown in barriers, ground and foliage. Figures 3 and 4. The calculations are performed with computer software that uses ray-tracing algorithms with optimized search algorithms for relevant rays from all sources to each calculation point, including direct rays, reflections, and diffractions. The model includes case specific: 1) emission 2) geo 3) building 4) barrier 5) ground 6) foliage data and 7)calculation parameters. Sound emission or source data includes position, orientation, directivity, level, emission spectra, or frequency-dependent sound power level data (complex transfer function, delay in filter settings). Geospatial data includes elevation points characterized by a 3- dimensional array of longitudinal, horizontal, and elevation data or isoheight lines of various resolution. Buildings are characterized by their geometry data (position, area and height), but also by their sound reflection properties. The barriers are characterized by their height, depth, position and shape as well as by their sound absorption or reflection properties. The ground area is determined by its Fig 3.: dBA and dBC maps using spectrum of moderate geometry and its absorption properties, which are defined music [7]. by a factor G. The foliage data is mainly determined by its area. The calculation parameters include: 1) calculation area, 2) grid resolution, 3) temperature, 4) humidity, 1st Čurović et al.: Development of a special standard for outdoor music events in Slovenia based on measurements and calculations 272 AAAA – 2023 – IZOLA - Conference Proceedings arrays emit coherent sound waves, and complex summation should be used instead of simple energy summation (Equation 2). Complex summation is not currently part of commercial software, which means that modelling with available software should be subjected to a validation procedure. 4. CONCLUSION In the first part of this work, the legislation in force in Slovenia was analysed and the following was established: - The radiation characteristics of the loudspeaker are determined by the data of the electrical power of the amplifier, which are not applicable in practical (real) situations and cannot be monitored by the people concerned or even by experts in sound engineering and acoustics. - The sound propagation model does not include all attenuation factors that are included in modern sound propagation models. - The directivity of loudspeakers is based on the mathematical theory of Lord Rayleigh from the 19th century, which does not take into account the Fig 4.: dBA and dBC maps using spectrum of electronic complexity of modern sound reinforcement systems. music [7]. - Prediction by calculation is done only for a single frequency, and the spectral characteristics of the radiated sound are not taken into account. - Monitoring of the ambient sound pressure level can be done by measurements; however, the integration time and measurement locations do not take into account the dynamics of the musical event and the signal-to-noise ratio in the far field. - The results of monitoring may not be known until after the event, and monitoring cannot be used to adjust the performance of the sound reinforcement system in real time. In the second part, an overview of the state of the art was shown. It was noted that a specialized standard for outdoor music events should cover several complex topics, including low-frequency noise, integration time, on-site and far-field monitoring, loudspeaker directivity, and sound emission spectra. An important step toward standardization of calculations would be a publication (a guide to good practice) of model parameters, including those for sound radiation, calculation, and propagation, which could then be used by acoustical experts in their studies, as is the practice in many acoustical standards. Because of the large number of influencing parameters, Fig 5.: dBA maps using different number of reflections in such a task can only be undertaken by a group of the calculations (1 vs 3) [7]. professionals that includes environmental noise specialists, sound engineers, event organizers, and With respect to sound propagation modelling, source regulators. To carry out such a project, a communication parameters should be carefully selected. The loudspeaker 1st Čurović et al.: Development of a special standard for outdoor music events in Slovenia based on measurements and calculations 273 AAAA – 2023 – IZOLA - Conference Proceedings specialist could be brought in to avoid communication [5.] Uredba o načinu uporabe zvočnih naprav, ki na errors between the various parties involved. shodih in prireditvah povzročajo hrup (Uradni list RS, št. 118/05 in 44/22 – ZVO-2). Available at: http://www.pisrs.si/Pis.web/pregledPredpisa?id=UR ED3652 (Accessed 01 September 2023) 4. REFERENCES [6.] Directive 2002/49/EC of the European Parliament and of the Council of 25 June 2002 Relating to the [1.] Environmental Noise Guidelines for the European Assessment and Management of Environmental Region, World Health Organization. Regional Office Noise - Declaration by the Commission in the for Europe, 2018. Conciliation Committee on the Directive Relating to [2.] Hill, AJ Understanding and managing sound the Assessment and Management of Environmental exposure and noise pollution at outdoor events, Noise. Available at: https://eur-lex.europa.eu/legal- AES, 2020. content/EN/TXT/?uri=celex%3A32002L0049 [3.] ISO 1996-2:2017. Acoustics – Description, (Accessed 20 January 2022) measurement and assessment of environmental [7.] d&b NoizCalc: Assessing and modelling noise noise — Part 2: Determination of sound pressure immissions in the far field. Available at: levels, International Organization for https://www.dbaudio.com/global/en/products/soft Standardization, 2017. ware/noizcalc/?pk_source=Newsletter&pk_medium [4.] ISO 9613-2:1996. Acoustics – Attenuation of sound =Dotmailer&dm_i=3WB1%2CGVZR%2C48ASA8%2C1 during propagation outdoors — Part 2: General T3IG%2C1 (Accessed 22 August 2023) method of calculation, International Organization for Standardization, 1996. 1st Čurović et al.: Development of a special standard for outdoor music events in Slovenia based on measurements and calculations 274 Contributed papers Advanced meausrement techniques in acoustics 1. Estimating Speed of Sound in Granular Materials: Impulse Response Extraction and Wave Decomposition in an Extended Impedance Tube Anže Železnik (University of Ljubljana, Faculty of Mechanical Engineering, LDSTA) 2. A smart method to calibrate universal testing machines by incorporating acoustic methods Sharath Peethambaran Subadra (Hochschule für Angewandte Wissenschaften Hamburg) 3. Unsupervised Classification of Welding processes based on Psychoacoustic Sound Features Jurij Prezelj (University of Ljubljana, Faculty of Mechanical Engineering) 275 Estimating Speed of Sound in Granular Materials: Impulse Response Extraction and Wave Decomposition in an Extended Impedance Tube Anže Železnik, Luka Čurović, Jure Murovec, Nejc Cerkovnik, Jurij Prezelj University of Ljubljana, Faculty of Mechanical Engineering Abstract: The speed of sound in materials greatly affects their acoustic properties and is a key factor in determining the acoustic impedance of a surface, which governs the reflection, absorption, and transmission of acoustic energy. Granular materials offer an intriguing alternative to custom-made acoustic materials, as they can achieve high sound absorption or insulation through the use of different size fractions. By investigating the speed of sound within the pores of granular materials, one can estimate their acoustic properties. In this paper, a method for measuring the speed of sound in porous materials using an impedance tube and time-domain wave decomposition of the tube's impulse response is proposed, employing a single downstream microphone to assess various size fractions of recycled sand. The measurements encompassed granular fractions ranging from 0.1 to 4 mm. Notably, decrease in the speed of sound within granular materials as the granules became smaller can be observed, with the speed dropping below 100 m/s for the smallest fraction. Keywords: impedance tube, time-domain wave decomposition, speed of sound, granular materials 1. INTRODUCTION properties of granular materials using a simple deconvolution-based method to separate sound waves in Granular materials have emerged as a promising avenue the time domain. Our approach is reminiscent of the in the field of sound absorption materials due to their "Adrienne" reflection method introduced by Garai [16-favourable acoustic properties and relatively low cost 18], but adapted for measurements within a closed tube. compared to traditional acoustic foams. Extensive Similar methodologies have been utilized by Sun et al. research has been conducted on the absorption [19,20] and Lefebvre [21] for sound absorption coefficient of various granular materials, including glass measurements, and subsequently modified by Sun et al. spheres [1], rubber crumb [2], and natural and recycled [22] for transmission loss measurements. The impedance materials [3-11]. However, the investigation of sound tube employed in our study is akin to the method insulation properties in granular materials remains proposed by Sun et al., albeit with sound pulses obtained relatively limited [12-14]. Nevertheless, existing findings through deconvolution rather than direct measurements suggest that utilizing very small granules can yield high using a pulse generator. transmission loss values. When examining the transmitted sound pulse in the Impedance tubes have been widely employed to study the impedance tube, a delay can be observed when acoustic properties of granular materials. However, the comparing the transmitted sound with and without a standard transfer matrix method faces challenges when sample inserted into the tube. This delay, previously measuring the transmission loss of highly reflective observed and employed by Oblak et al. [23] for speed of samples like granular materials [13,15]. To overcome sound measurements in porous materials submerged in a these limitations, we constructed an extended impedance liquid using the transfer matrix method [24], provides tube specifically designed to examine the sound insulation valuable insights into the influence of pore size on the Železnik et al.: Estimating Speed of Sound in Granular Materials: Impulse Response Extraction and Wave Decomposition in an Extended Impedance Tube 276 AAAA – 2023 – IZOLA - Conference Proceedings speed of sound within porous media. Because the speed frequency range of the impedance tube is limited by the of sound in a porous material influences the surface tube diameter that determines the highest frequency that impedance, it is an important factor in determining can be measured due to cross-modes and by the signal-to- acoustic properties of materials. Granular materials, due noise ratio and frequency resolution at lower frequencies. to their adjustable particle and pore sizes achievable The valid frequency range of the impedance tube used in through different size fractions, offer an intriguing avenue this paper is between 100 and 4000 Hz. A schematic of the for investigating this phenomenon. For this study, measurement setup is shown in Fig. 1. recycled silica sand was separated into size fractions ranging from 0.1 mm to 3 mm using laboratory test sieves and a shaker. Two methods were evaluated to measure the delay in the impulse response of the transmitted sound wave caused by the granular material sample. The first method involved direct calculation of the delay in the impulse response, while the second method utilized phase differences in the phase of the measured impulse responses to derive a frequency-dependent speed of sound characteristic. To ensure accurate measurements of the speed of sound, multiple thicknesses of granular material samples were used to achieve a high signal-to- noise ratio. 2. METHODS The impedance tube used for the measurements in this paper consisted of two long tube sections (4 m and 2 m) with a sample holder between them. The use of a long impedance tube enables the separation of sound waves Fig.1. Schematic of the measurement setup. coming from different directions in the time domain. A single microphone placed behind the sample was used to The equipment used for the impedance tube: obtain impulse responses of the impedance tube with and without samples inserted. Samples were sealed in the • Brüel & Kjær Sine Random Generator Type 1027 sample holder using a dense sealing paste to reduce signal generator flanking sound transmission. Signals from a signal • Brüel & Kjær Power Amplifier Type 2706 power generator and a measurement microphone were used to amplifier deconvolve the impulse response of the impedance tube • Visaton R 10 S loudspeaker using spectral division and a 6-th order Butterworth filter • PCB Piezotronics 130F20 microphone was used to eliminate all frequencies outside the • Data Translation DT9847-3-1 A/D converter frequency range of interest. A sampling frequency of 192 kHz was used for the recordings to eliminate aliasing in the The microphone behind the sample can be used to frequency range of interest and custom python software measure the transmitted sound before and after a sample was used to calculate the speed of sound within samples. is inserted into the tube. The impulse response with and Because spectral division was used for deconvolution, without a sample inserted is shown in Fig. 2. Multiple sound recordings with the duration of 60 seconds were reflections can be observed in the impulse response, used and no time averaging was employed. The speaker including the reflection off the tube termination. Because and microphone were mounted according to the of that, the impulse response needs to be time-windowed recommendations found in EN ISO 10534-2:2001. The to include only the directly transmitted sound. This enables the measurement of transmission loss using this Železnik et al.: Estimating Speed of Sound in Granular Materials: Impulse Response Extraction and Wave Decomposition in an Extended Impedance Tube 277 AAAA – 2023 – IZOLA - Conference Proceedings type of time-domain wave separation as a reduction of In this paper, two methods are compared for determining amplitude is evident when a sample is inserted (Fig. 2 b)). the speed of sound in porous samples: using only the delay in the maximum value of the impulse response (as shown in Fig. 3) and using the phase difference in the impulse responses to determine the speed of sound for the entire valid frequency range of the impedance tube. For both methods, the speed of sound cs is calculated using the measured delay caused by the sample ( Δt), the thickness of the sample ( ds) and the reference speed of sound in air ( c0). 𝑑 𝑐 𝑠 𝑠 = (1) 𝑑𝑠 + 𝛥𝑡 𝑐0 For the method using the phase of the impulse responses, cs is obtained by calculating the phase of both impulse responses from their Fourier transform, unwrapping the phase and calculating the time delay between the phase characteristics for each frequency. This method allows for Fig.2. The impulse response without a sample inserted the evaluation of the calculated values and proved (a) and with a sample inserted into the tube (b). effective for eliminating measurements where the signal- to-noise ratio was not sufficient because of the high The speed of sound inside a porous material can be transmission loss values of samples and the method using measured inside an impedance tube because a change in only the delay between the two impulse responses was the speed of sound inside a sample causes a delay in the not sufficient. transmitted sound wave. If the impulse responses without The two main sources of measurement errors for speed of and with a sample are compared, this delay can be sound measurements are the sampling frequency and the observed. The comparison between the reference signal-to-noise ratio. The limited sampling frequency impulse response and an impulse response obtained with causes errors for measurements where the speed of a granular material sample is shown in Fig. 3. sound is not much different compared to the reference speed of sound and the impulse response is only shifted for a few samples. This is evident for granular fractions that do not change the speed of sound drastically, especially for thin samples. Because of this, thicker samples need to be used for measurements where the speed of sound within samples is similar to the speed of sound in air. Signal-to-noise ratio causes measurement errors where the transmission loss values of samples are high. This means that thinner samples yield more reliable results for samples where the speed of sound is very low (and transmission loss values are high). The samples for the speed of sound measurements were prepared with a set of laboratory test sieves and a shaker. Ten different size fractions were measured: 0-150 μm, 150-180 μm, 180-210 μm, 210-250 μm, 250-300 μm, 300- 425 μm, 425-600 μm, 600-1000 μm and 3-4 mm. To Fig.3. The delay caused by the sample. ensure a sufficient signal-to-noise ratio for the Železnik et al.: Estimating Speed of Sound in Granular Materials: Impulse Response Extraction and Wave Decomposition in an Extended Impedance Tube 278 AAAA – 2023 – IZOLA - Conference Proceedings measurements, 7 different thicknesses were measured insufficient). Examples for three granule size fractions are for all the fractions: 5 mm, 10 mm, 20 mm, 30 mm, 40 mm, shown in Fig. 5. Frequency ranges and sample thicknesses 60 mm and 80 mm. where speed of sound measurements are deemed valid are shown in green. The three size fractions shown are 3- 4 mm (a), 425-600 μm (b) and 150-180 μm (c). While the 3. RESULTS AND DISCUSSION thicker samples are used for larger size fractions (a) and most sample thicknesses are appropriate for some The average measured speed of sound using only the granular size fractions (b), only the thinnest samples are delay in the maximum value of the deconvolved impulse appropriate for measuring the speed of sound of the response for all sample thicknesses is shown in Fig. 4. The smallest granules (c) as the transmission loss of these error bars indicate the standard deviation for samples is very high and the signal-to-noise ratio is measurements of the same granule size fraction. It is insufficient. evident that the standard deviation using this method is too high to reliably estimate the speed of sound in granular samples. Because there is no way to estimate which measurements are valid in terms of signal-to-noise ratio with this method, using only the delay in the impulse response is deemed insufficient for reliable speed of sound measurements in porous samples. Fig.4. Measured speed of sound using the delay in deconvolved impulse responses. Using the method comparing the phase delay in the impulse responses gives a frequency dependent speed of sound characteristic that can be used to estimate the separate measurements and the frequency ranges in which the speed of sound measurements are valid. For this study, the criteria for the speed of sound measurements to be considered valid are: a) the speed of sound characteristic is relatively flat across the frequency range of interest (eliminating resonances and measurements where the signal-to-noise ratio is insufficient) and b) speed of sound measurements of multiple sample thicknesses have similar values Fig.5. Valid frequency ranges and sample thicknesses: a) (eliminating measurements where phase delay is 3-4 mm, b) 425-600 μm and c) 150-180 μm granules. Železnik et al.: Estimating Speed of Sound in Granular Materials: Impulse Response Extraction and Wave Decomposition in an Extended Impedance Tube 279 AAAA – 2023 – IZOLA - Conference Proceedings The averaged speed of sound values meeting the Fig. 7 b). The flow resistivity measured without granular predefined criteria were computed. These mean values, material samples was 300 Pa s m-2, and the fitted curve along with their corresponding standard deviations, are intersected this value at around 380 m/s. This intersection depicted in Fig. 6. Evidently, the speed of sound within provides a relatively accurate estimation of the speed of granular material pores is notably lower as the granule sound in air, given the multitude of parameters size decreases. This phenomenon contributes to the high influencing measurement outcomes. transmission loss values observed in granular materials [14]. This occurrence can be attributed to the substantial surface impedance mismatch arising from the considerably lower speed of sound within air-saturated granular materials compared to that in air. Consequently, a significant portion of sound is reflected from the granular material surface. The lower speed of sound in smaller granules could be attributed to increased sound scattering across numerous surfaces within the porous material, coupled with the longer trajectory traversed by the transmitted sound within smaller granular fractions. Fig.7. The relation between speed of sound and Fig.6. Measured speed of sound using the phase measured flow resistivity. difference in deconvolved impulse responses. While the speed of sound in granular materials decreases 4. CONCLUSION for smaller granules, the flow resistivity of smaller granular fractions increases significantly. Static flow In this paper, a new approach for measuring the speed of resistivity values for 40 mm thick samples were measured sound in porous media using phase differences in using the direct flow method ( ISO 9053-1:2018). Steady deconvolved impulse responses using a long impedance airflow was generated using a diaphragm pump and a 20 l tube is presented. The objective of this study was to pressure reservoir. A volumetric air flow meter and a develop a simple method that can be used for fast and pressure gage were used for measuring air flow and the reliable measurements of speed of sound while also pressure difference. Samples were mounted in the measuring transmission loss values. The measurement apparatus inside the same containers used for impedance results confirmed that the new method using phase tube measurements and were sealed with a dense sealing differences is more reliable compared to the method paste. The comparison between the trends for speed of using the delays in impulse responses. The results of sound and flow resistivity values dependent on particle measurements of different granule sizes show that the size is shown in Fig. 7 a) and the relation between the speed of sound is much lower for smaller granules, under measured speed of sound and flow resistivity is shown in 100 m/s for particles with diameters around 100 μm. Železnik et al.: Estimating Speed of Sound in Granular Materials: Impulse Response Extraction and Wave Decomposition in an Extended Impedance Tube 280 AAAA – 2023 – IZOLA - Conference Proceedings 5. REFERENCES [13.] G. Pispola, K. V. Horoshenkov, F. Asdrubali, Transmission loss measurement of consolidated [1.] S. Sakamoto, Y. Sakuma, K. Yanagimoto, S. Watanabe, granular media (L), The Journal of the Acoustical Basic study for the acoustic characteristics of granular Society of America 117 (5) (2005) 2716-2719. material (Normal incidence absorption coefficient for [14.] A. Železnik, J. Murovec, L. Čurović, N. Cerkovnik, multilayer with different grain diameters), Journal of J. Prezelj, Transmission loss measurement of recycled Environment and Engineering 7 (1) (2012) 12-22. granular material using wave decomposition by [2.] M.J. Swift, P. Bris, K.V. Horoshenkov, Acoustic impulse response extraction based on deconvolution, absorption in re-cycled rubber granulate, Applied Applied Acoustics 211 (2023) 109498. Acoustics 57 (3) (1999) 203-212. [15.] D. Caballol, Á.P. Raposo, Analysis of the [3.] F. Asdrubali, K.V. Horoshenkov, The acoustic measurement of transmission loss in rigid building properties of expanded clay granulates, Building materials with a standing wave tube, Construction Acoustics 9 (2) (2002) 85-98. and Building Materials 182 (2018) 242-248. [4.] J. Sikora, J. Turkiewicz, Sound absorption coefficients [16.] M. Garai, Measurement of the sound-absorption of granular materials, Mechanics and control 29 (3) coefficient in situ: the reflection method using (2010) 149-157. periodic pseudo-random sequences of maximum [5.] H. Zhou, B. Li, G. Huang, Sound absorption length, Applied acoustics 39 (1-2) (1993) 119-139. characteristics of polymer microparticles, Journal of [17.] P. Guidorzi, M. Garai, Advancements in sound Applied Polymer Science 101 (4) (2006) 2675-2679. reflection and airborne sound insulation [6.] S.H. Zolanvari, C. Karlstetter, Development of measurement on noise barriers, Open Journal of Consolidated Granular Materials as Sound Acoustics 3 (2013) 25-38. Absorbers: Acoustical and Mechanical [18.] M. Garai, P. Guidorzi, European methodology for Characteristics, in: Proc. AIA-DAGA, 2013. testing the airborne sound insulation characteristics [7.] S. Sakamoto, M. Sasaki, I. Kourakata, K. Yanagimoto, of noise barriers in situ: Experimental verification and S. Watanabe, Basic study for acoustic absorption comparison with laboratory data, The Journal of the characteristics of soft and light granular material Acoustical Society of America 108 (3) (2000) 1054- (basic characteristics for expanded polystyrene 1067. beads), Journal of Advanced Mechanical Design, [19.] L. Sun, H. Hou, L.Y. Dong, F.R. Wan, Sound Systems, and Manufacturing 7 (4) (2013) 677-689. absorption measurement in a circular tube using the [8.] A. Biskupičová, M. Ledererová, S. Unčík, C. Glorieux, echo-pulse method, Acta Acustica united with M. Rychtáriková, Sound absorption properties of Acustica 96 (5) (2010) 973-976. materials based on recycled plastic granule mixtures, [20.] L. Sun, H. Hou, Measurement of sound Slovak Journal of Civil Engineering, 29 (1) (2021) 15- absorption by underwater acoustic material using 19. pulse-separation method, Applied Acoustics 85 [9.] C. Bujoreanu, F. Nedeff, M. Benchea, M. Agop, (2014) 106-110. Experimental and theoretical considerations on [21.] A. Lefebvre, G.P. Scavone, J. Abel, A. Buckiewicz- sound absorption performance of waste materials Smith, A comparison of impedance measurements including the effect of backing plates, Applied using one and two microphones, in: Proceedings of Acoustics 119 (2017) 88-93. ISMA, 2007. [10.] A. Boubel, M. Garoum, A. Bybi, N. Laaroussi, [22.] L. Sun, H. Hou, Transmission loss measurement of Sound absorption measurements of some porous acoustic material using time-domain pulse- loose granular materials, in: INTER-NOISE and NOISE- separation method (L), The Journal of the Acoustical CON Congress and Conference Proceedings, 2016. Society of America 129 (4) (2011) 1681-1684. [11.] S. Sakamoto, K. Yamaguchi, K. Ii, R. Takakura, Y. [23.] M. Oblak, M. Pirnat, M. Boltežar, An impedance Nakamura, R. Suzuki, Theoretical and experiment tube submerged in a liquid for the low-frequency analysis on the sound absorption characteristics of a transmission-loss measurement of a porous material, layer of fine lightweight powder, The Journal of the Applied Acoustics 139 (2018) 203-212. Acoustical Society of America 146 (4) (2019) 2253- [24.] B. H. Song, J. S. Bolton, A transfer-matrix 2262. approach for estimating the characteristic impedance [12.] E. Levi, S. Sgarbi, E. A. Piana, Acoustic and wave numbers of limp and rigid porous materials, Characterization of Some Steel Industry Waste The Journal of the Acoustical Society of America 107 Materials, Appl. Sci. 11 (2021) 5924–5924. (3) (2000) 1131-1152. Železnik et al.: Estimating Speed of Sound in Granular Materials: Impulse Response Extraction and Wave Decomposition in an Extended Impedance Tube 281 A SMART METHOD TO CALIBRATE UNIVERSAL TESTING MACHINES BY INCORPORATING ACOUSTIC METHODS Sharath P. Subadra*, Shahram Sheikhi* *Institute of Materials Science and Joining Technology, Research and Transfer Centre 3i; Hamburg University of Applied Science, Berliner Tor 13, 20099 Hamburg, Germany Abstract: The aim of this research is the development of a non-destructive testing methodology for tensile, compression and bending tests to ascertain the calibration status of universal testing machines. An algorithm which would be eventually included in the machine-bound control system keeps track of the calibration status by assessing errors which would be dependent on the acoustically determined material properties like elastic, shear and compression moduli. An "Acoustic Reference Sample" serving as a non-destructive reference material would be used to determine these properties. The reference sample material would have a large elastic range and hence can be loaded within this range several times to obtain the elastic properties, which would be subsequently cross-referenced with the acoustically determined material properties in the calibration stage. The machine-bound control system would make use of an inbuilt algorithm to keep track of this change in material properties and hence the calibration status. The development of this strategy would bring down costs associated with quality assessments and subsequently improve reliability of testing procedures considerably when there are long intervals in calibration of the testing equipment. Keywords: Non-destructive testing, material properties, acoustics, calibration, universal testing machine, programming 1. INTRODUCTION [2]. Dynamic mechanical analysis is a powerful technique Elastic properties of materials especially the Youngś where the sample is forced to vibrate in its natural modulus (E-modulus or simply E) is an important material frequency [5]. These frequencies can then be plugged into property, and it plays crucial role in design considerations relevant equations to obtain the elastic properties such as and material quality control. These properties are Youngs and shear modulus [6,7]. Impulse excitation certified according to DIN EN 10204 material test technique (IET) is a widely used technique to identify the certificates. The material tests to determine the E- natural frequencies. The method involves measuring the modulus can be either a direct or an indirect method, oscillations introduced by the impulse which can be where the former involves the application of longitudinal measured by microphone, contactless vibrometer or a and transverse deformations, whereas the latter involve piezo crystal [8]. deducing the E from other elastic constants such as shear However, there exists differences in the E measured from and bulk moduli respectively [1,2]. Direct methods static and dynamic tests respectively. Trubitz et al. [9] typically involve measuring the E from the slope of a compared the estimated E from static and dynamic tests stress-strain curve in the linear region obtained after a for various grey cast iron materials, and it was observed tension test, but the tests though widely used is expensive that ultrasonics tests gave 5% higher E values vis-à-vis a [3]. The material properties obtained via direct test statically determined test such as a uni-axial tensile test. methods are subject to calibration errors in the Similarly, Peixoto et al. saw 2% higher values when equipment and the inherent methods adopted while comparing a dynamic method with that of static methods, testing. Indirect methods are much more adaptable to here the comparison was between one dynamic and two industrial use and are easier, less expensive, and less time- static methods respectively. consuming [4]. Dynamic measurements to determine E is As opposed to the dynamic methods elaborated in the much more accurate and the error is in the order of ± 1% previous paragraphs, the elastic modulus obtained 1st Author Surname et al.: Paper title 282 AAAA – 2023 – IZOLA - Conference Proceedings through tensile testing methods as per relevant standards acronym ZIM. A reference sample is being developed as (EN 10002-1, ASTM E8 / E8M-09 and EN ISO 6892-2) part of this project which can be used in conjunction with exhibits scatter and variations. But these methods are a non-destructive method (acoustic method) on a advantageous owing to their wide usage and the continuous basis without destroying the sample. The possibilities of determining all the subsequent material reference sample is ideally loaded in the lower elastic properties, e.g. the beginning of yield, ultimate strength region to measure the modulus of elasticity and shear etc [10]. Therefore, these two methods (dynamic and modulus, and these properties are compared with static) can complement each other in calibration efforts acoustically determined (dynamic measurement) moduli. especially when universal testing machines (UTMs) are If there are deviations from the measured value, the considered. Since, UTMs are still the norm in extracting machine has deviated from its ideal calibrated state, since material properties, the reliability of the results depends the dynamically determined parameters are dependent on the calibration status of the UTM. Normally, the on the geometry and boundary conditions of the calibration is carried out once a year and in many areas of reference sample when its natural frequencies are application this is not sufficient, since a significant excited. In addition to the non-destructive reference variation in the calibration leads to faulty batches. This is sample, an algorithm is being developed which is intended true especially in the automotive, aviation and railway to perform calibration in two stages which would later be sectors where non-conformity after shipment can lead to incorporated into the machine interface along with costly claims of damages. The calibration errors in UTMs relevant hardware such as sensors and pre-amplifiers. can occur on account of faulty software, an inclined position of the crosshead, faulty grips etc[1-10,13]. 1.1. EXPERIMENTAL SETUP AND CALIBRATION A method that is in practise in the context of calibration of ALGORITHM UTMs and impact testing machines is the use of reference 1.1.1 Theoretical considerations samples that are certified and conforming to the Determination of E of a prismatic beam with an respective standards. The material parameters of these asymmetric cross-section around one axis is based on reference samples are known which can be used to three vibration modes namely the out-of-plane and in- calibrate the equipment. Therefore, the certified plane bending modes respectively along with longitudinal reference samples can provide indirect proof of stability, vibration modes. Out-of-plane modes are easy to excite accuracy and precision of the test equipment. By referring and there is minimal risk of combined loads appearing the reference sample parameters, the suitability of the [20]. Bending vibration of beams obeying the Eulers- setting parameters in the testing software can be checked Bernoulli theory is a better choice to ascertain the modes as well. Reference samples also form a tool for standard- and the respective frequencies. Since, the derivations of compliant determination of the measurement un- the equations for various boundary conditions are widely certainty which is of interests to the customers. Large available, hence the derivations are beyond the scope of intervals associated with this calibration procedure means this paper. The equation (below) stands true for a beam that the equipment is not monitored continuously, and simply supported, fixed ends and a cantilever type hence significant deviations are often not taken into the (boundary conditions). The coefficient 𝑪𝒏 takes into these account. Using reference samples which are irreversibly boundary conditions and the mode shapes (table 1). The destroyed during calibration process adds to the expenses other terms in the equation are either geometric associated with this calibration procedure[1-10, 13]. parameters or material properties of the structure and its This paper is part of a research project titled material respectively. I is the moment of inertia of the “ReuseDetect” carried out at Hochschule für Angewandte cross-section, h is the thickness, 𝝆 the density, L the length Wissenschaften Hamburg in cooperation with of the structure and f the natural frequency [11]. Schütz+Licht Prüftechnik GmbH, wherein the aim is to develop a non-destructive measurement technique to 𝑬𝑰 𝝎 ≈ 𝑪 (1) continuously monitor the calibration status of a UTM. The 𝒏 = 𝑪𝒏√𝝆𝑨𝑳𝟒 𝒏√ 𝑬𝒉𝟐 𝟏𝟐𝝆𝑳𝟒 project is funded by the Federal Ministry of Economic 𝝆𝑳𝟒 𝟒𝟖𝝅𝟐𝒇𝟐 𝑬 = ( ) (2) 𝒉𝟐 𝑪𝟐 Affairs and Climate Action under the aegis of Central 𝒏 Innovation Program for SMEs known by its German 1st Author Surname et al.: Paper title 283 AAAA – 2023 – IZOLA - Conference Proceedings Table. 1. 𝑪𝒏 for different boundary conditions and modes Table. 2. Chemical composition of the steel alloy [11] Chemical Composition [%] Material Boundary Conditions Ti Nb V Modes (n) Simply Fixed-Fixed Fixed-Free Steel 0.18 0.10 0.08 Supported Tensile tests were performed on a QASAR 200 tensile 1 𝝅𝟐 22,3733 3,5160 testing machine from Galdabini (Cardano al Campo, Italy). 2 𝟒𝝅𝟐 61,6728 22,0345 The machine was in a fully calibrated stage, and hence 3 would serve as an equipment on which the algorithm 𝟗𝝅𝟐 120,9034 61,6972 developed would be put into test. The maximum load cell 4 𝟏𝟔𝝅𝟐 199,8594 120,0902 capacity of this machine is 200 kN, the grippings are 5 𝟐𝟓𝝅𝟐 298,5555 199,8600 hydraulic driven, while the loading is servo motor driven. Acoustic signals were collected from a QASS Optimizer 4D 1.1.2 Materials, equipments and design of experiments system from Qass GmbH (Wetter-Ruhr, Germany). The A non-destructive reference sample was developed for entire test setup can be seen in fig 2, were the Qass the purpose of calibration (fig 1-A) from a high strength system is equipped with a 4 MHz sensor along with pre- low alloy steel commercially available as ALFORM700M® amplifiers. The sampling rate was fixed at 400000 Hz for (termed as M700 in this paper), the chemical the purpose of signal conversion from time domain to compositions are seen in table 2. The sample dimensions frequency domain. were machined as per DIN 52125. A typical micro- structure of this steel is shown in fig 1-B (cross-section of the sample), which is dominated by granular bainite ferrite and martensite [12]. Fig.2. The test setup along with tensile testing machine and Qass acoustic system. Two set of tests were performed at a strain rate of 0.00007/s and 0.00025/s respectively. The maximum yeild strenght of M700 was 781 MPa and hence all the tests were stopped at 550 MPa, this is to ensure that the same sample can be used several times for the purpose of calibration. Once, the stress-strain data has been obtained for these strain rates, a regression model was implemented on python to obtain the best fitting line from the stress-strain plot and hence the Young’s modulus (E). 25 tests were perfromed at each of the strain rates and the E was tabulated for each of the tests to measure the standard deviation and hence to understand the reliability of the material as a universal non-destructive Fig.1. A) Reference sample (DIN 52125), B) Grain reference sample. For the sake of consistency 3 sample structure (cross section of the sample) of ALFORM700M® where considered for the study. 1st Author Surname et al.: Paper title 284 AAAA – 2023 – IZOLA - Conference Proceedings Acoustic signals are introduced into the sample, were the from calibration Youngs modulus along sample is constrained in a manner as seen in fig 3. Since, certificate (data for with data from the project is in a stage of progression, different methods load cell and strain frequency spectrum, where tried, like introducing sound in the form of impulse gauge) while uncertainty data directly into the sample by hitting it, introducing sound is used to measure through the machine (by knicking at positions where deviation of UTMs singal is not lost in the form of damping etc), or by using a calibration speaker and playing loud music. Signals were obtained 3 Perform FFT on the Generating the before and after loading (after 25 loadings) to look into signal data frequency spectrum to any form of deviations. The sampling rate was fixed at estimate the resonant 400000 Hz. With the precise boundary conditions known frequency and fixed, the Youngs modulu’s is calculated using 4 Looking for the The resonant equation (2). The entire calibration process is elaborated resonant frequency frequencies are used in section 1.1.3 in the spectrum, by to estimate the comparing the Young’s modulus peaks in the based on equations spectrum with the elaborated in eq (2) frequencies estimated earlier in step 2. 5 Performing As the machine loses machine learning its calibration the Fig.3. Typical reference sample contrained in a cantilever approach based on uncertainty differs, manner to excite the flexural modes. regression which the algorithm modelling to find reads from the Table. 3. Material properties of M700 Youngs modulus measured data and Material properties M700 from stress-strain adds up to the Youngs Youngs Modulus, E [GPa] 206 data. modulus calculated Yield at 0,2% strain [MP] 781 and if this total (Emax) Ultimate Strenght [MPa] 800 exceeds exceeds 3% [13] of the first 1.1.3 Algorithm for calibration of the universal testing measured E-modulus machine of the reference The algorithm is shown in the form of action taken by the sample then the user and corresponding execution to be carried out in the machine has partially table below (Table 4). lost its calibration. 6 Comparing the The E-modulus Table. 4. Actions taken by the user and execution by the Youngs modulus measured from algorithm measured from acoustic is given a max Step Action Result acoustic to that and min differing by 1 Importing the The signal is in CSV measured from 3%. If this range is still signal (time format and is imported tensions test. within the Emax from domain), and by the algorithm, and step 5, the machine is stress-strain date the signal is shown in still in calibration. But from Qasar 200 fig 4 (A). if E-modulus from step 2 Input specimen The specimens 5 exceeds or is less dimensions, dimensions are used than Emax and Emin in uncertainty data to determine the 1st Author Surname et al.: Paper title 285 AAAA – 2023 – IZOLA - Conference Proceedings this step, the machine on the data obtained from Qasar 200, to obtain the best has lost its calibration. fit for curve. The E-modulus was measured within a range of the elastic region of the stress-strain plot and the procedure elaborated in ISO 6892 was adopted. In 2. RESULTS AND DISCUSSIONS keeping with the norm, an upper and lower stress values 2.1 Frequency spectrum analysis were identified where the upper limit corresponds to 40% The algorithm reads the signal data generated and of Rp0.2 while the lower limit was set at 10-20% of Rp0.2. performs a fast fourier transforms to convert the time Within this region a linear regression is performed via domain signal into a frequency domain singal also termed machine learning tools to obtain a curve fit. Coefficient of as a spectrum. The algorithm check the spectrum for determination R2 would serve as reminder of determining domianant flexural modes, and latter untilises it for how good the curve fit is. In this case it was as close as calcuting the E-modulus. The algorithm knows the 0.9992(this is close as recommended in per ISO 6892-1: frequencies based on a data-base it has access to having 2019). Table 5(a) and table 5(b) shows the measurement E-modulus of M700 steel from literatur. It calculates the from sample 1 at strain rates 0.00007/s and 0.00025/s frequenices based on the E-modulus and looks for similar respectively. It must be stated here that each of these frequencies within the spectrum. This is performed from samples were loaded 25 times each at the respective step 1 to step 4 in table 4. A typical spectrum is seen in fig strain rates. The plots for the same sample can be seen in 4-B. fig 4(a) and (b). A closer look into the plots reveals that at lower rates of loading, the plots seem to wither away towards the end of the test, while at a slightly higher rate of loading, the plots were coincident to a large extend and hence the standard deviation in E-modulus measurements were low. Thus, based on this non- destructive loading within the elastic region it was concluded that, for an effective calibration to be carried, the strain rate shall be fixed at 0.00025/s. The process elaborated in this section shall be performed from stage 5 until 6 in table 4. Table. 5(a). E-modulus measurement at 0.00007/s Test No Strain [%] E [GPa] 1 0.27 201.68 2 0.27 209.56 3 0.27 210.77 4 0.27 207.77 5 0.27 207.97 St.Dev 1.415 Table. 5(b). E-modulus measurement at 0.00025/s Test No Strain [%] E [GPa] 1 0.27 207.36 Fig. 4. A typical acoustic signal introduced by impulse 2 0.27 207.88 excitation. A) Time-domain signal, B) Frequency domain 3 0.27 207.32 signal. 4 0.27 206.38 5 0.27 207.25 2.2 E-modulus measurement from Qasar 200 St.Dev 0.539 Test were performed at two different strain rates to ascertain the correct rate which could be used to perform the calibration. Thus, regression modelling was performed 1st Author Surname et al.: Paper title 286 AAAA – 2023 – IZOLA - Conference Proceedings high strength low alloy steel commercially available as ALFORM700M® was used. Acoustic methods where initially implemented to obtain the Young’s modulus, latter the sample was strained within the elastic region to obtain Young’s modulus. Considering all the uncertainties stemming from the strain gauge, load cell etc, the E- Fig. 4. Stress strain plot for single sample loaded at two modulus was tabulated. An algorithm was implemented different rates (A) 0.00007/s, (B)0.00025/s. to determine the modulus from these two methods and 2.3 Calibration monitoring of Qasar 200 same was latter used to ascertain the calibration status of As mentioned in table 4, the algorithm read both the the universal testing machine. Several more rounds of signal data from acoustics and the data from Qasar 200. testing would be implemented before, the algorithm Initially it generates a signal plot in time domain, the data would be integrated into the machine’s interface so that a is latter sliced if necessary, based on the amount of data. live monitoring of the equipment calibration status can be An FFT is performed on it and later the spectrum is implemented. generated, and from the spectrum natural frequencies in flexural modes are identified and its used to tabulate the 4. REFERENCES E-modulus. A max and min is assigned to this calculated E- [1.] Sonja, K. et al. Uncertainty in the determination of modulus termed as Emax and Emin, whereas the modulus elastic modulus by tensile testing, Engineering calculated from dominant frequencies is termed as E-aco. Science and Technology, an International Journal, 25, And based on the E-modulus comparisons from the 100998, 2022. tension test, the calibration is monitored. [2.] Viala, R., Placet, V., and Cogan, S. Identification of the anisotropic elastic and damping properties of complex shape composite parts using an inverse method based on finite element model updating and 3D velocity fields measurement (FEmu-3DFV): application to bio-based composite violin soundboards, Compos. Part A Appl. Sci. Manuf., 106, pp. 91-103, 2018. [3.] Sebastián, T., Walter, S., Alberto, S., and Angel, M. Measurement of the Youngś modulus in particulate epoxy composites using the impulse excitation technique, Materials Science and Engineering A, 527 pp. 4619 – 4623, 2010. [4.] Lord, J.D., Morrell, R.M. Elastic Modulus Measurement, Good Practice Guide No. 98, National Laboratory, 2006. [5.] Ho-Cheung Ho. et al. Mechanical properties of high strength S690 steel welded sections through tensile tests on heat-treated coupons, Journal of Constructional Steel Research, 166, 105922, 2020. [6.] Advanced Technical Ceramics-Mechanical Properties of Monolithic Ceramics at Room Temperature-Part 2:Determination of Youngś Modulus, Shear Modulus and Poissonś Ratio; German Version EN 843-2:2006; Beuth Publishing DIN: Berlin, Germany, 2007. Fig. 5. A flow-chart of the algorithm for calibration [7.] Philipp, L., Georg, F., Christoph, H., Florian, S., Florian, E., and Wolfram, V. Acoustical and optical 3. CONCLUSION determination of mechanical properties of This paper was an attempt to use non-destructive testing inorganically-bound foundry core materials, techniques to ascertain the calibration and possible Materials, 13(11), 2531. [8.] Roebben, G., Bollen, B., Brebels, A., van Humbeeck, calibration of universal testing machine. Within this J., van der Biest, O. Impulse excitation apparatus to context a non-destructive test specimen machined from a measure resonant frequencies, elastic moduli, and 1st Author Surname et al.: Paper title 287 AAAA – 2023 – IZOLA - Conference Proceedings internal friction at room and high temperature, Rev. Sci. Instrum, 68, pg. 4511-4515, 1997. [9.] Trubitz, P., Rehmer, B., Pusch, G. In: Die Ermittlung elastischer konstanten von gusseisenwerkstoffen. Tagung wekstoffprüffung, Neu-Ulm, Germany, pp. 267-272. [10.] Sebastian, S., and Marion, M. A new approach for the determination of the linear elastic modulus from uniaxial tensile tests of sheet metals, J. of Mat. Processing Tech., 241, pg. 64-72, 2017 [11.] Barboni, L., Gillich, G.R., Chioncel, C.P., Hamat, C.O., and Mituletu, I.C. A method to precise determine the Youngsś modulus from dynamic measurements, In: Conf. Ser.: Mater. Sci. Eng, 416 012063. [12.] Klein, M., Spindler, H., Luger, A., Rauch, R., Stiazny, P., Eigelsberger, M. Thermomechanically hot rolled high and ultra high strength stell grades- processing, properties and application, Material Science Forum. Trans Tech Publications, Ltd., November 2005. [13.] Lord, J.D., Rides, M., Loveday, M.S. TENSTAND WP3 final report: modulus measurements methods. NPL Report. DEPC-MPE 016. URI: http://eprintspublications.npl.co.uk/id/eprint/3223 1st Author Surname et al.: Paper title 288 Unsupervised Classification of Welding processes based on Psychoacoustic Sound Features Jurij Prezelj, Jaka Lavrih, Damjan Klobčar Faculty of mechanical engineering, University of Ljubljana Abstract: Welding processes, to ensure the quality of the final products, often need to be controlled through external sensors. In MAG/MIG welding, the welder or welding operator also relies on hearing, i.e., audible sound emissions during the welding process. From the interpreted sound emissions during the welding process, the stability of the welding process can be determined. In the paper, we analysed short-circuit and pulsed MIG/MAG welding under different operating conditions. Signals of welding current, voltage and sound in the audible and ultrasonic range were recorded. During welding, we simulated poor welding conditions such as lack of protective gas atmosphere, distancing the welding torch from the welding surface, and improper preparation of the workpieces with greasy, coloured, and oxidized surfaces. Additionally, we applied unsupervised K-Means classification in combination with psychoacoustic features of audible sound to check the hypothesis that sound emission carries enough information about the process to classify it. By changing the number of classes, it was found out that sound can be used to monitor the welding process, identify the welding process type and further it revealed a specific mode of welding. Keywords: Welding sound, Audible noise of welding, Psychoacoustic features, Unsupervised classification, K-Means 1. INTRODUCTION welding process, it's also necessary to know other data about the environment, such as the local shape of the All areas of the industry are becoming increasingly weld joint with potential irregularities, the temperature of optimized or refined, including welding. Real-time welding the workpiece, the state of the workpiece clamping and process optimization is already being used in modern its distortion, sound, etc. [4]. By knowing the appropriate welding processes. Adjusting the process in real-time values, we can adjust the joining process to ensure the allows for improved quality, reduced production times, appropriate changes to welding parameters and and cost savings due to the elimination of defects, consequently achieve the desired quality of the welding reworks, and inferior products. The outcome of the process and product. manufacturing process in the industry are often various products that must be made within appropriate 1.2 Goals tolerances and quality. The purpose of the study was to experimentally check 1.1 Motivation the influence of disturbances on the welding result and sound emissions during the weld surfacing, at which In automated welding and joining processes, there is a welding process disturbances occurred. A welding desire to use various sensors to control the joining process experimental plan was designed and executed to obtain in real-time and thus prevent the production of inferior information about the course of the process from various products [1, 2]. In arc welding processes, welding current sensors. Based on the measured values of electrical sources allow the setting of welding parameters, and the quantities, sound quantities, and visual assessment, the controller of the source adjusts welding parameters based aim was to determine the correlations between input and on electrical quantities [3]. To ensure the quality of the output quantities and induced errors during welding. 1st Author Surname et al.: Paper title 289 AAAA – 2023 – IZOLA - Conference Proceedings During experimental welding, we induced typical errors and disturbances that usually occur during the welding or weld surfacing process. We simulated these disturbances and errors with greasy stains, protective paint coatings, excessive surface oxidation (corrosion), and lack of a protective atmosphere, representing drafts. The goal of the task was to determine the correlation between process parameters and measured sound quantities in the ultrasonic range and whether the measured quantities could be used in real-time optimization. Fig. 2: Schematic representation of spherical disturbance propagation 2. THEORETHICAL BACKGROUND Calculation of the pressure depending on the radius r In welding, we assume that the ignition of the welding arc as a function of volume flow, generated by the oscillations is a monopole source of sound wave propagation with a of the arc volume: changing volume of the medium. In the MIG/MAG 𝑟 𝜕𝑄 (𝑡 − ) process, when the welding wire contacts the workpiece, a 𝜌 𝑝(𝑟, 𝑡) = 0 𝑐 [𝑃𝑎] (1) short circuit occurs, leading to the formation of the arc. 4𝜋𝑟 𝜕𝑡 where: r is the distance to the microphone and r/c is 2.1 Generation of sound during welding the time delay, Q(t) is the fluid volume flow generated by air displacement around the oscillating arc, The arc ignition heats the surrounding gases, which increase in volume. When the formed droplet at the end (0) of the welding wire detaches, the arc extinguishes, the 𝑄(𝑡) = ∯ 𝑣(𝑡, 𝑟, 𝜙)𝑑𝑆 gases cool down, and their volume decreases. The volume's rate of change over time gives us the arc surface Equation for calculating sound pressure change acceleration. clarifies that sound pressure depends on power supplied to the electric arc, i.e., its volume: 𝑑 𝑝(𝑡) = 𝐶1 (𝑈𝐼) (3) 𝑑𝑡 where C1 is the arc geometry constant and depends on Fig. 1: Creation of air pressure disturbances during r, U is arc voltage, and I is welding current intensity. Arc welding arc ignition voltage is calculated using the equation [4, 5]: These changes result in air pressure variations or 𝑑𝐼 𝑅𝑎𝐼 pressure disturbances, which we call the sound source. 𝑈 = 𝑅𝐼 + 𝐿 + (4) 𝑑𝑡 1 + 𝑖𝜔𝐶𝑎𝑅𝑎 The formation of the arc and the creation of pressure wave disturbances are shown in Figure 1. The sound where: I stands for electric current intensity, R a is arc pressure is calculated according to Eq.1, related to Figure resistance, R c is arc capacitance, and ω is angular 2. frequency. Eq. 4 therefore indicates that there is a direct The connection between the electric current during correlation between welding current and sound pressure. welding and created sound pressure p disturbances is Figure 3 represents the link on how welding current presented in equation Eq.3 and in Fig.3, which is graphical signals affect the generated sound signal. Each pulse of illustration of Eq.3. electric current can be linked to a change in sound pressure. The welding current increases during the electrode's short circuit with the workpiece, after which the welding arc ignites. After the arc ignition, with some 1st Author Surname et al.: Paper title 290 AAAA – 2023 – IZOLA - Conference Proceedings delay, the shock wave of the surrounding gases reaches In this equation, ETQ is the excitation at threshold in quiet our microphone, which we denote as time delay td. and E0 is the excitation that corresponds to the reference Audible sound of welding has been studied and since intensity I0 = 1012 W/m2. The specific loudness reaches the direct correlation to welding power has been proven, asymptotically the value N0 = 0 for small values of E. it was used for real time welding monitoring. So far, no Loudness N is then the integral of the specific loudness studies have been done on application of psychoacoustic over the critical-band rate z in [Bark], or in the features for real time welding classification, although mathematical expression: psychoacoustic features prooved to be very useful in combination with unsupervised classification algorithms, (6) [8, 9, 10, 11] 2.2.2 Roughness Roughness correlates to how noticeable or annoying a Arc ignition sound is as heard by the human ear [12]. More specifically, Short circuit roughness is a hearing sensation related to loudness Arc extinction modulations at frequencies too high to be discerned Welding current separately, such as modulation frequencies greater than 30 Hz. The roughness R in [asper] of any sound can be calculated using the following equations: Acoustic signal Time delay td (7) Time delay td (8) 0 2,08 4,16 6,25 8,33 10,42 12,50 14,6 16,67 18,75 20,83 time /msec/ In Eq. (8), Nmax and Nmin are the maximum and minimum Fig. 3: Relationship between the generation of the specific loudness in the current critical band, f welding arc and the change in air pressure mod is the modulation frequency and LE(z) is the amplitude of 2.2 Psychoacoustic metrics and signal descriptors modulation. Psychoacoustic metrics, like loudness, roughness, 2.2.3 Sharpness sharpness, and tonality were extracted, from the recorded Sharpness corresponds to the sensation of a sharp, sound signals, together with some other signal features painful, high-frequency sound and it represents the like number of pulses, which represent number of short comparison of the amount of high frequency energy to the circuits events during welding. This feature also proved to total energy [12]. It is calculated as a weighted area of be important during the classification process. loudness, like an area moment calculation, as shown in Eq. (9). In Eq. (9), g(z) is the weighting function that has a 2.2.1 Loudness unitary value of 1 below 3 kHz and non-linearly increases Loudness is a term referring to the human perception of from 3 kHz to 20 kHz, where it has a value of four. For high sound level [12]. The definition of loudness states that 1 values of sharpness, significant spectral components at sone corresponds to a 1 kHz tone at 40 dB. The loudness high frequencies are necessary. scale quantifies loudness to the human ear. Loudness represents the dominant feature for the evaluation of (9) sound quality. To calculate it, specific loudness has to be determined, as given in Eq. (5). 2.2.4 Tonality Tonality represents the auditory perception character (5) related to the pitch strength of sounds. There are many models for calculating tonality; in this paper, the Aures model is used [12]. where q1 (zi), q2 (fi) and q3 (Li) represent weighting functions based on the bandwidth 1st Author Surname et al.: Paper title 291 AAAA – 2023 – IZOLA - Conference Proceedings (Eq. (11)), centre frequency (Eq. (12)) and prominence (Eq. K 2 k arg min        x − = arg min       c Varc i i i i (15) (13)) of each tonal component and NGr is the loudness of i=1 x c  i= i 1 C C the broadband noise component. Tonality is used to determine whether a sound consists mainly of tonal where ei is the average of points attributed to ci. This is components or broadband noise [12]. equivalent to minimizing the pairwise squared deviations of points in the same cluster: (10) 2 K 1      k x − (11) arg min y (16) i 1 = 2c i x, yc i C (12) The equivalence can be deduced from identity: k 2 (13)  x−ε = (x−ε i)(ε −y i ) (15) x c  xy c  i i 2.3 Unsupervised k-means classification algorithm Because the total variance is constant, this corresponds to maximizing the sum of the squared The k-means algorithm is an iterative algorithm that deviations between points in different clusters, which attempts to classify the data set into K different, non- follows from the law of total variance. The k-means overlapping clusters, where each data point can only algorithm is usually initiated with a randomly filled matrix belong to one group. Data points are D-dimensional of centroids C. Euclidean distances are then calculated for vectors, with each component representing a feature each pair of data points to each centroid representing a extracted from observations. The k-means algorithm class. Each data point is then assigned to a class, whose attempts to arrange N observations in K clusters so that distance to the centroid is minimal. The new centroid is the data points within the clusters are adjacent, while calculated from the new distribution of data points for keeping the clusters as far away from each other as each class. The new centroid of a given class is an possible. The evaluation is based on the Euclidean averaged value of all data points assigned to this class. The distance between the points, [8,10]. The distance Euclidean distances are calculated again and processed between a centroid with the index c continuously until there are no more transitions of data m and the observed point x points between the classes, [8,10]. n is defined by an equation, D 3. EXPERIMNTAL SETUP 2 c x = c 2 x 2 m n  − m i, n i, (14) i=1 The set of devices for conducting experiments consisted of a welding current source for performing K-means algorithm assigns each data point to the welding, a 3-axis CNC positioning system for moving the nearest cluster defined with its centroid. The algorithm is weldment, measuring instruments for capturing electrical based on minimizing the arithmetic mean of all data quantities, and measuring instruments for capturing points that belong to the same cluster. The smaller the sound quantities. The entire system is shown in Figure 4. variation within clusters, the more homogeneous the data A 3-axis scissor type positioning system was used for points are within the same cluster. K-means clustering movement, providing linear movements in three different therefore aims to classify the N observations into K (≤ N) axes. It was made in the Laboratory for Technical clusters defined with the centroids C = {c1, c2, …, ck} to Cybernetics, Machining Systems and Computer minimize the sum of the squares within the cluster. Technology LAKOS, at the Faculty of Mechanical Formally, the objective is to find: Engineering of the University of Ljubljana. It was later converted into a manipulator for 3D welding. A protective container was mounted on the movable table of the 1st Author Surname et al.: Paper title 292 AAAA – 2023 – IZOLA - Conference Proceedings positioning system, which protects the machine and the electrode was on the negative pole of the welding table. environment from UV radiation and molten metal When measuring arc voltage, a voltage divider with a splashes during the welding process. The clamping surface factor of 10 was additionally connected to the circuit. The was air-cooled and electrically isolated from other parts of measuring card allows us to measure voltage up to 10 V, the machine. It is shown in Figure 4, [3]. which is beyond the operating level of the welding source. d) Sound pressure For measuring sound pressure, a system of a microphone, microphone amplifier, and measuring card was used. The microphone was of the Bruel & Kjaer 4138 type. It was connected to the Bruel & Kjaer 2636 type microphone preamplifier, from which the measured values were stored on the computer through the measuring card. Figure 4.9 shows the used microphone, and Figure 4.7 shows the used microphone pre-amplifier. The measured data was displayed in the LabView software environment after the experiments were Fig. 4: Layout of the experimental system for welding completed. For the calculations of sound quantities, a 1 Welding current source Daihen Varstroj P500L LabView program named 'Feature Extraction' was used, 2 Feeder unit with additional material which was developed in the Laboratory for Technical 3 Mechanism of the X positioning system Acoustics of the Faculty of Mechanical Engineering of the 4 Protective container with weldment University of Ljubljana. The results were displayed and 5 Computer with LabView software compared in certain frequency ranges of audible sound 6 Computer for controlling the CNC system with G-code and ultrasound. The effective sound pressure value (RMS 7 Microphone pre-amplifier B&K 2636 8 Gas cylinder with protective gas Ferroline C18 of sound), roughness, and fluctuation intensity were 9 Microphone for capturing sound waves B&K 4138 displayed in the frequency range between 80 kHz and 120 kHz for the entire length of the weld. Additionally, the a) Measurement card NI 9222 recorded sound was processed with a low-pass and high- All signals were aquired with a measuring card National pass filter in the frequency range between 200 Hz and 80 Instruments NI 9222 with the corresponding NI cDAQ- kHz. Low-pass filtration was used due to disturbances 9174 housing. It is shown in Figure 4. Through the from the controlled source and high-pass filtration due to measuring card, measurements of electric current, environmental disturbances where the experiments were electrical voltage, and sound pressure were recorded. For conducted. measuring electrical current, we used additional output signal from the current clamps. For measuring electrical voltage, the measuring electrodes were connected in 4. EXPERIMENTAL PLAN parallel to the welding process circuit. b) Electric current The aim of the experimental work was to demonstrate the It was measured indirectly through the current clamps. influence of disturbances during the welding process on The current clamps were from Voltcraft, type VC-511, the quality of the weld and its effect on the measured placed on the negative pole of the welding source. results. Modern welding sources have the ability to Through their voltage output, they were connected in overcome negative impacts during welding due to their parallel to the measuring card. The output voltage factor advancement. Despite poor conditions, they can in some against the measured electric current of the current clamp instances still provide a high-quality welding result. Before is 1 mV/A. the execution of the experiments, a test plan was c) Electrical voltage prepared. The plan was specified all the welding One electrode was connected to the positive pole inside parameters, the type of disturbance to be executed during the feeder unit of the welding source, and the other welding, and the required preparation of the base surface 1st Author Surname et al.: Paper title 293 AAAA – 2023 – IZOLA - Conference Proceedings to be welded on. The test plan was made for easier The chosen welding methods were Short Circuit (DC Low monitoring of the test execution, fewer chances of Spatter) and Pulsed (DC Pulse). The aim was to compare mistakes, and the most efficient use of time and work. It the two advanced modes of the MIG/MAG welding source is presented in Table 1. operation. The Short Circuit with the Low Spatter function, The chosen welding methods were Short Circuit (DC Low as the name suggests, reduces spattering of the molten Spatter) and Pulsed (DC Pulse). The aim was to compare material during welding. Pulsed welding adds electrical the two advanced modes of the MIG/MAG welding source current pulses of approximately 400 A to the welding operation. The Short Circuit with the Low Spatter function, current signal for better defined detachment of the as the name suggests, reduces spattering of the molten material droplets. The welding parameters were set on material during welding. Pulsed welding adds electrical the front control panel of the welding source. Table 1 current pulses of approximately 400 A to the welding displays the settings for each test. The welding torch's current signal for better defined detachment of the travel speed was the same for all tests, namely 300 material droplets. The welding parameters were set on mm/min. It was set during the trajectory programming of the front control panel of the welding source. Table 1 the Positioning System X. The initial distance of the displays the settings for each test. The welding torch's welding torch, or the length of the free wire end, was 10 travel speed was the same for all tests, namely 300 mm. mm/min. It was set during the trajectory programming of the Positioning System X. The initial distance of the The welding trajectory was programmed in a program welding torch, or the length of the free wire end, was 10 created in the LabView software environment. Other mm. parameters related to the movements of the 3-axis CNC system X were also set in this program. The data was then Table 1: Welding test execution plan, welding current 90A. transferred to the AXIS computer program, which controls Weld Welding Distortion Gas Voltage the CNC manipulator. The trajectory of each weld is shown No. type flow [V] in Figure 5. The length of each weld was 145 mm, and the [l/min] distance of the free electrode end increased along its 1.1 (DC Reference weld 13 21,8 entire length with seven equal increments of 2.5 mm. A pulse) transition can be seen between each increment. The 1.2 (DC low Reference weld 13 15,3 purpose of increasing the torch distance was to capture spatter) the influence of different heights during a single test. 1.5 (DC No protective 0 21,8 pulse) atmosphere 1.6 (DC low No protective 0 15,3 spatter) atmosphere 1.7 (DC Protective 13 21,8 pulse) paint layer on the surface 1.8 (DC low Protective 13 15,3 spatter) paint layer on Figure 5: Welding torch trajectory for welding seams. the surface Used Materials 2.1 (DC low Oily surface 13 15,3 spatter) The base plate onto which the welding was done is made 2.2 (DC Oily surface 13 21,8 of structural steel labeled S235 J0. The dimensions of the pulse) plate were 140 × 200 × 5 mm. Structural steel is most used 2.6 (DC Corroded 13 21,8 in construction. It has good weldability, formability, and pulse) welding surface mechanical properties. The material of the base plate was chosen due to its widespread use in the industry and 2.7 (DC low Corroded 13 15,3 availability. The physical properties of the material are: spatter) welding surface Density 7.85 [g/cm3]; Linear expansion coefficient 1st Author Surname et al.: Paper title 294 AAAA – 2023 – IZOLA - Conference Proceedings 1.08·10-5 [K-1]; Thermal conductivity 56.3 [E/m·K]. For each type of sample, an individual base plate was Chemical composition of S235 J0 steel [5]: C =0.17%, Mn prepared for welding. For reference samples and those =1.40%, P=0.035%, N=0.012, Cu=0.55 and remaining is Fe. lacking a protective atmosphere, the surface of the base Mechanical properties of S235 J0 steel [21]: Yield strength plate was only cleaned with a solvent. For samples welded Rp0.2 = 235 [MPa], Tensile strength Rm = 360-510 [MPa], on an oily surface, a thick layer of NLGI-2 type machine Toughness KV > 27 [J] at 0°C. grease manufactured by Olma was applied to the welding The filler material was structural steel in the form of area. The corroded surface was pre-treated with a wire labeled VAC60, manufactured by SIJ Elektrode solution of hydrogen peroxide and water. A cloth soaked Jesenice, with a diameter of 1.2 mm. It was chosen due to in the hydrogen peroxide solution was placed on the its frequent use in industry and compatibility with the cleaned, degreased surface for 24 hours. After halting the base plate material onto which the welding was done. Its chemical reaction, the solution's remnants were washed chemical composition: C=0,06-0,1%, Si=0,9%, Mn=1,5%, off the plate. The plate with a protective paint layer was S<0,025%, P<0,025% and Fe in remaining percentage. The degreased and then given two passes with an hour mechanical properties of additive material: Yield strength interval of a thin layer of spray paint. Prepared welding Rp0.2 > 240 [MPa], Tensile strength Rm 500-640 [MPa], plates are shown in Table 2. Toughness KV >47 [J] at -50°C. [6] The shielding gas used in all tests was a gas mixture for 5. RESULTS MAG welding, specifically labeled Ferroline C18 produced by Messer. The gas composition was 18% CO2, and 82% For each weld signals were recorded, and signal Ar. Gas flow rates were 13 l/min for optimal conditions features were extracted. Examples of extracted and 0 l/min to demonstrate the case without a protective psychoacoustic features are presented in Figs. 6a and 6b. atmosphere. According to the manufacturer's data, the shielding gas is intended for welding plain, low-alloyed, and fine-grained structural steels by the MAG process. [10] Table 2: Prepared base plates for welding Unprepared base a) plate surface Surface prepared with a protective layer of paint b) Figure 6: Psychoacoustic features and corresponding welds for done with DC low spatter a) referenced weld Surface prepared welded at 150 A and b) weld welded at 90 A without the with an shielding gas. application of machine grease Each set of extracted values from the signal corresponds to the aligned location on the weld. Each Prepared corroded weld was performed under different conditions, surface consequently different noise was emitted. Difference can be observed by different values of signal features and consequently in their feature vector combinations of the psychoacoustic features. From Fig. 6a and 6b can also be 1st Author Surname et al.: Paper title 295 AAAA – 2023 – IZOLA - Conference Proceedings clearly visible that any errors during welding can be easily process is interrupted, or in other words, when the arc detected. This means that sound can be easily analysed goes out. during the process and used for the classification of the welding process itself. Statistical analysis of distribution of normalized feature values is presented in Fig. 7. From this figure different psychoacoustic features have different responses to defferent welding conditions. Only tonality is not relevant, becouse its value is for almost all samples always close to zero. Other psihoacoustic features Sharpness, Fluctuation strength and Roughness have faverable distribution. Fig.8: Classification of welding into two classes In classifying into three categories, the algorithm distinguished elements into idle elements when the process is not operating, and onto elements it had previously identified as the welding process. Furthermore, it divided them into two additional categories. It turned out that the algorithm separated the process elements based on the basic welding mode, that is, elements belonging to the DC Pulsed welding mode and those Figure 7: Statistical distribution of psihoacoustic features pertaining to Short Circuit DC Spatter welding. Table 3 displays the Confusion matrix. We observe that for DC Unsupervised classification is a statistical tool where an pulsed welding, the process is interrupted due to algorithm categorizes elements into classes. Each element disturbances more frequently (64 times), whereas, in the has its own set of features, which mathematically can be DC Spatter welding, it only stops 19 times. The algorithm represented by a vector. In our case, an element is a independently recognized that it could not classify the portion of the weld created over a duration of 1 second. elements into either welding method but identified them From the measured signals within this 1-second period, as interruptions. we extract feature values. In our example, we used psychoacoustic features from the sound signal. The dimension of the vector is equal to the number of features. For the classification procedure, we utilized the K-means algorithm, often mentioned as a component of artificial intelligence. With unsupervised algorithms, we don't train the system; instead, we allow the algorithm to learn from the dataset on how to classify elements. The basic algorithm we used cannot determine the number of classes on its own, so we conducted the classification for three predetermined numbers of classes; 2, 3, and 4. Fig.9: Classification of welding into three classes In classifying elements into two classes, the algorithm recognized the difference between elements belonging to From the confusion matrix in table 3, it can be inferred the welding process and those related to idling, i.e., when that the algorithm mistakenly identified DC pulsed the welding process is not operating, as depicted in Figure welding as DC spatter welding only twice, and 8. From the image, it is clear that the algorithm can detect misidentified short circuit DC welding as DC pulsed when a welding disturbance is so significant that the welding just eight times. Such misidentifications occurred 1st Author Surname et al.: Paper title 296 AAAA – 2023 – IZOLA - Conference Proceedings only under extremely adverse conditions when the welding process was highly unstable. This indicates that the algorithm ensures more than a 95% accuracy rate. Table 3: Confusion matrix for classification K-Means Class (0) (1) (2) No welding DC Pulsed Spatter DC Pulsed 64 171 2 (1) Short circuit DC 19 8 289 Spatter(2) Fig.11: Classification of welding into five classes Next, we increased the number of categories to four. The algorithm split category 2, which belongs to the DC 3. CONCLUSION spatter welding method. It classified 5 welds into an additional category, which are indicated in the image Unsupervised classification of welding, based on below, Figure 10, with arrows. From our analysis of psychoacoustic features of the sound generated by unsupervised classification, it turned out that the welding, has revealed that sound can be easily used to algorithm automatically identified welds that were monitor the welding process. Furthermore, the welded with an increased current of 150 A. The algorithm experiment demonstrated that different welding methods thus independently recognized welding with increased can be differentiated based on classification with current as a new welding method. psychoacoustic sound features. The K-Means algorithm correctly distinguished between DC sputter and DC pulsed welding methods with a probability of more than 95%. Additionally, the use of unsupervised classification proved to be a useful tool for identifying processes that might otherwise be hidden in the noise of feature values. Through unsupervised classification, we identified that the activity during DC Spatter welding at very high currents over 150 A is intense, and the material transfer mechanism is changed, which can be easily identified by using classification algorithm based on psychoacoustic features. Fig.10: Classification of welding into four classes 4. REFERENCES When increasing the number of categories to five, the algorithm divided category 1, representing DC Pulsed, into [1.] Zhang YM. Real-Time Weld Process Monitoring. Cambridge: Elsevier Science; 2008. two subcategories. In Figure 11, it's clear which welds http://qut.eblib.com.au/patron/FullRecord.aspx? from group 1 were newly categorized. However, as the p=1639763 number of categories increased, the clarity of [2.] Kah, P., Layus, P., Hiltunen, E., Martikainen, J., classification became blurred. This is because the welding 2014. Real-Time Weld Process Monitoring. AMR conditions within a single weld, due to the inclination of 933, 117– the weld piece and the consequent increasing distance 124.https://doi.org/10.4028/www.scientific.net/a between the weldment and the electrode, varied. As a mr.933.117 [3.] Ščetinec A., Klobčar D., Bračun D. In-process path result, the algorithm began to distinguish categories replanning and online layer height control within a single weldment. through deposition arc current for gas metal arc based additive manufacturing, Journal of 297 AAAA – 2023 – IZOLA - Conference Proceedings Manufacturing Processes, 64, pp. 1169-1179, [8.] Murovec, J., et.al., Automated identification and 2021, assessment of environmental noise sources, https://doi.org/10.1016/j.jmapro.2021.02.038 Heliyon, Volume 9, Issue 1, January 2023, e12846 [4.] Prezelj, J., et.al., Noise as a signal for on-line [9.] Murovec, J. et.al., Psychoacoustic approach for estimation and monitoring of welding process. cavitation detection in centrifugal pumps, Applied Acta acustica united with Acustica. 2003, letn. 89, Acoustics, Volume 165, August 2020, 107323 št. 2, str. 280-286. [10.] Prezelj, J., et.al., Identification of different [5.] Horvat, J., et.al., Monitoring Gas Metal Arc manifestations of nonlinear stick–slip phenomena Welding Process by Using Audible Sound Signal, during creep groan braking noise by using the Strojniški vestnik - Journal of Mechanical unsupervised learning algorithms k-means and Engineering, Vol. 57, (2011)3, pp. 267-278 self-organizing map, Mechanical Systems and [6.] Major, M., Nawrot, J., and Major, I. Structural S235 Signal Processing, Volume 166, 1 March 2022, and S355 Steels – Numerical Analysis of Selected 108349 Rods Connection, IOP Conference Series: Materials [11.] Lipar, P., et.al., Psychoacoustic approach used for Science and Engineering, Volume 585, 5th Annual developing the model of sound pleasantness of International Workshop on Materials Science and vacuum cleaners and suction units based on Engineering 17–18 May 2019, Hunan, Changsha, objective and subjective analysis, 5th Congress of China, pp. 2, 2019. Alps-Adria Acoustics Association, pp. 12-14 [7.] Sij elektrode: welding consumables. Available at: [12.] Hugo Fastl , Eberhard Zwicker, Psychoacoustics https://www.sij.si/assets/magazinefiles/Elektrod Facts and Models, Springer Berlin, Heidelberg, W eldinConsumables.compressed.pdf (Accessed 2007 10 August 2023). 298 Index A Acoustic environment modeling, 26 Acoustic heritage, 205 Acoustic performance, 35, 36, 37, 76, 103, 181, 182, 202 Acoustic scanning, 148 Acoustical features, 26 Acoustics laboratory, 175, 244, 245 Active learning, 142, 143, 145, 146 Advanced signal processing methods, 16 Alpha chamber, 11, 17, 39 Ambisonics, 14, 68, 168, 207, 238, 239, 240, 241, 242, 243, 244, 245, 248 Audible noise of welding, 292 Audio forensics, 26 Audiometry, 209, 215, 216, 217 Authenticity verification, 26 Axial fan, 76, 77, 79, 80, 81, 82, 83 B Binaural audio, 202 C Calibration, 79, 92, 95, 112, 131, 136, 138, 139, 216, 228, 229, 236, 266, 285, 286, 287, 288, 289, 290 Centrifugal fan, 11, 75, 103, 116 Chambers, 41, 148, 152, 175 Classifier architecture, 26 Classifier performance, 26 Computer program, 154, 162, 298 Concert hall acoustics, 18 Control valve, 232, 236 Convolutional neural network, 26, 33 Criminal proceeding, 209, 217 D Dataset selection, 142 Diffuse sound field, 39 Digital image correlation, 16 Doors, 13, 127, 176, 177, 179, 180, 245, 250 E Electronical commutation, 84, 87 Energy labelling, 84, 85, 89 Environmental noise, 130, 132, 139, 140, 141, 263, 268, 272 Environmental performance, 35, 37, 182, 186, 189 Equipment noise, 11, 17, 53, 64, 66, 183 F Firecracker noise, 209, 217 Floating floors, 13, 176, 190, 192, 202 Flow induced noise, 232, 236 Ful y connected network, 26, 27, 33 299 G Granular materials, 277, 282, 284 H Harmonic timbre coordinates, 118, 125 Head tracking, 202 Health, 76, 115, 127, 130, 139, 154, 155, 175, 183, 186, 209, 211, 212, 213, 214, 216, 219, 220, 221, 222, 227, 230, 271 Hearing loss, 154, 209, 214, 217, 218, 219, 220, 221 High impulse noise, 209, 217 Higher order ambisonics, 244 High-speed camera, 6, 16 Household appliance, 11, 75, 103 Human perception, 6, 31, 77, 115, 167, 222, 239, 294 Human rights violation, 208, 209, 217 I Image-based methods, 16 Immission directivity, 130, 132, 135 Impact Noise, 14, 238, 239 Impact sound, 13, 74, 151, 170, 176, 183, 184, 185, 186, 187, 189, 190, 192, 193, 194, 196, 198, 199, 200, 202, 239, 240, 243, 244, 245, 247, 248, 261 Impact sound insulation, 13, 176, 184 Impedance tube, 40, 277, 279, 280, 282 Impeller geometry, 83, 84 Individual hrtfs, 13, 204, 202 Infrasound, 227, 231 Interactive material, 164 L Labview, 39, 40, 227, 229, 230, 231 LCA study, 35 Linear timbre vector space, 118 Listening experiment, 202 Loudness, 54, 57, 77, 80, 108, 109, 111, 113, 114, 115, 162, 172, 294, 295 Loudspeakers, 170, 173, 244, 245, 247, 268, 270, 271, 272, 274 Low frequency noise, 227, 228, 231 M Mass-spring system, 172, 190, 192, 193 Measurement automatisation, 148 Microphone cable robot, 148 Microphone differential array, 130 Mobile application, 127, 128, 130 Motorways, 250, 257 Multichannel Software, 39 Music therapy, 154 Musical instruments, 118, 122, 125, 126 N Neural network, 26, 27, 28, 32, 33, 142, 143, 147 Noise barriers, 250 Noise emission model, 106, 109, 101 Noise generation mechanism, 84 Noise mapping, 130, 137, 138, 168 Noise propagation model, 98, 104 300 Noise source position, 64, 68, 69, 71, 73 Non-destructive testing, 285 O Online learning, 164 Open courseware, 164 Outdoor events, 268, 271, 275 Overal noise assessments, 116 P Pass-by noise measurements, 98, 100, 105 Passenger cabin, 13, 208, 227 Perception evaluation, 127 Plausibility, 244, 245 Primary excitation, 237 Pseudosound, 11, 75, 103 Psychoacoustic features, 76, 108, 292, 299 Psychoacoustic indicators, 115 Psychoacoustics, 7, 76, 77, 83, 118, 164, 167, 173 Psychoaoustics, 11, 75, 103 R Rail roughness, 98, 100, 101 Railway, 11, 75, 98, 99, 100, 101, 102, 104, 105, 106, 101, 138, 250, 259, 286 Railway noise, 98 Rehabilitation, 154, 155 Resonance frequency, 107, 108, 109, 171, 190, 192, 193, 196 Resonant Absorption, 39 Reusing waste foam, 35 Reverberation time, 20, 21, 22, 24, 39, 41, 42, 43, 44, 45, 46, 47, 50, 51, 52, 56, 62, 64, 66, 68, 69, 108, 109, 148, 152, 163, 168, 171, 179, 183, 241 Reverberation time, 11, 17, 39, 46, 62 Room acoustics, 18, 33, 40, 64, 170, 185 Room EQ Wizard, 227, 228, 231 Room impulse responses, 26 Rotor mass unbalance, 84 S Sensor network, 127 Sound absorbers, 35 Sound absorption Coefficient, 39 Sound essays, 13, 204, 205, 207, 209 Sound event detection, 142, 145, 146, 147 Sound insulation, 6, 13, 35, 44, 64, 66, 74, 151, 170, 175, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 192, 193, 194, 196, 198, 199, 200, 202, 218, 229, 243, 248, 258, 261, 266, 267, 277, 284 Sound quality metrics, 6, 82, 115 Sound quality metrics, 115 Soundscape, 13, 83, 115, 127, 130, 131, 168, 204, 205, 206, 201 Soundscapes, 205, 207 Spatial frequency, 39 Spectral optical flow imaging, 16 Spectrogram data, 26 Speech intel igibility, 11, 17, 64, 66, 68, 69, 71, 73, 74, 154, 156, 167 Speech rate, 52, 54 Speech signal, 52, 53, 57, 60, 155, 156 Stage design, 18 Steel railway bridges, 106, 101 Subglottal pressure, 52, 53, 54, 55, 60, 61 Sustainable building assessment, 182 301 T Test facility, 13, 149, 152, 176, 177, 178, 179, 180, 187 Timber constructions, 175 Time-domain wave decomposition, 277 Time-varying loudness models, 115 Tinnitus, 154, 155, 161, 162, 218, 219, 220 Track decay rate, 98, 100, 105 Turbulent flow, 11, 75, 103 Two-dimensional Fourier transformation, 39 U Uncertainty, 14, 42, 97, 123, 124, 131, 143, 204, 249, 258, 259, 260, 261, 263, 264, 265, 266 Uncertainty standard deviations, 258 Underlay materials, 13, 176, 190, 192, 193, 194, 196, 199 Universal testing machine, 285, 288, 290 University campuses, 13, 204, 205, 206, 201 Unsupervised classification, 292, 300, 301 V Vacuum cleaner suction unit, 75, 84, 85, 86, 87 Vandalism, 142, 143, 145 Vibration dampers, 11, 75, 106, 107, 108, 109, 101 Vibration reconstruction, 16 Vibroacoustic measurements, 98 Virtual-source approach, 6, 237 W Welding sound, 292 Well-being, 76, 115, 183 Wind turbine noise, 75, 91, 93, 94 302 Gold sponsors Silver sponsors Bronze sponsors