Modeling and experimental evaluation of CPU-GPU crossover points in OpenCL-based image processing

slovenian

2026

volume 93, issue 3

crossover point   GPU acceleration   image processing   OpenCL   performance modeling

Elektrotehniška zveza Slovenije
(Obvezni izvod spletne publikacije)

A novel approach is proposed to reduce the noise in cochlear implants by utilizing an enhanced multiband spectral subtraction method. The main objective is to optimize the subtraction factors used in the spectral subtraction process to improve the overall performance of the stimulation strategy in cochlear implants designed to decompose the incoming audio signals into multiple sub-bands for a further processing. However, one of the persistent challenges in the cochlear implant technology is maintaining the speech intelligibility in noisy environments. To solve the issue, the subtraction factors are dynamically adapted based on common noise patterns observed in real-world scenarios. The method effectiveness is assessed by conducting extensive simulations using noisy sentences corrupted by eight distinct types of the environmental noise, including the street noise, machinery, and crowd chatter. The objective evaluations are performed by using the established metrics for the speech intelligibility and audio quality. The results demonstrate that the proposed method significantly enhances the signal-to-noise ratio,thus ensuring are improved speech clarity for users of cochlear implants. The results obtained in our simulations show that the approach is highly promising. The more precise signal processing enabled improves the speech intelligibility and assures a user satisfaction.

20.10.2026