U][)C811.163.6'342.4H Peter Jurgec Fran Ramovš Institute of the Slovenian Language, Ljubljana FORMANT FREQUENCIES OF VOWELS IN TONAL AND NON-TONAL STANDARD SLOVENIAN The article presents formant frequencies of Standard Slovenian (SS) vowels as spoken by five tonal and five non-tonal speakers in citation form. The results and subsequent analysis of variance indicate two types of differences between both groups. In the tonal SS, [+ ATR] mid vowels have higher F1, and short [a] has considerably lower F1. Secondly, acute, circumflex, and short vowels of all phonemes are more dispersed in the tonal SS, the differences being statistically significant in most cases. This is a by-product of fundamental frequency and intensity distinctions in the two tones, and of duration/centralization effects in quantity contrast. These phenomena do not occur in the non-tonal SS. V članku so predstavljene formantne frekvence samoglasnikov standardne slovenščine, kot jih govori pet tonemskih in pet netonemskih govorcev v izoliranih besedah. Rezultati in statistična analiza kažejo na dve vrsti razlik med obema skupinama: (1) pri tonemskih govorcih imata srednja visoka samoglasnika višji F1, kratki [a] pa precej nižjega (je centraliziran). (2) Pri tonemskih govorcih se akutirani, cirkumflektirani in kratki samoglasniki posameznega fonema v večini primerov statistično različni. V akustičnem smislu je to predvsem posledica razlik v osnovni frekvenci in jakosti, deloma pa tudi trajanja oz. fonetične redukcije. Tega v netonemski standardni slovenščini ni. Key words: acoustic phonetics, formant frequencies, suprasegmentals, tone, Slovenian Ključne besede: akustična fonetika, formanti, formantne frekvence, nadsegmentne lastnosti, ton, tonem, slovenščina 1 Introduction1 Phonetic studies of lexical tones in pitch-accented languages usually include acoustic analyses of fundamental frequency, intensity (or amplitude), duration, and phonation types. Spectral characteristics, most prominently formant frequencies, are considered non-significant or only marginally affected, and thus left aside, when tone is in question. On the other hand, formant frequencies, formant bandwidths, and spectral balance are the primary indicators of vowel quality (e.g., correspondence between openness and F1), and also prone to phonological and phonetic influence of stress (cf. Sluijter and Van Heuven 1996). The dependence of formant frequencies on vowel duration, phonetic reduction, or undershoot effect, speaking rate and style (e.g., Lind- 1 The author wishes to thank Vesna Mildner, Mateja Blas and the speakers for their valuable contributions to this work. Any remaining errors are the author's. Earlier versions of the article (or parts thereof) have been presented at Between Stress and Tone Conference in Leiden (June 16-18, 2005) and the International Conference of Language Variation in Europe in Amsterdam (June 23-25, 2005). The ZRCola font, used in this text, was developed by Peter Weiss at The Scientific Research Centre of the Slovenian Academy of Sciences and Arts in Ljubljana (http://www.zrc-sazu.si). blom 1963, Gay 1978, Tuller idr. 1982, Miller 1989, Engestrand 1988, Bakran 1989, Fourakis 1991, Van Son and Pols 1992, Moon and Lindblom 1994, Fourakis idr. 1999, Pitermann 2000, Erickson 2002, and Jurgec 2005c, for Slovenian), speaker's gender and fundamental frequency (Murry and Singh 1980, Assmann and Nearey 1987, Childers and Wu 1991, Wu and Childers 1991, Simpson 2001, and Jurgec 2005b) have been researched extensively. Moreover, studies of formant frequencies in pitch-accented languages usually represent each prosodic combination individually, cf., vowel charts of Croatian in Bakran 1989, or Lehiste and Ivic 1963: 84. In the present study however, the interaction between tonal features (i.e., phonological features primarily encoded as fundamental frequency oscillations) and form-ant frequencies is addressed. The hypothesis is that in tonal languages, formant frequencies can be affected by tonal differences to a certain degree. This can be viewed primarily as a by-product of fundamental frequency and intensity. In respect to tonal features, Slovenian has two types of dialects, pitch-accented2 and stress-accented, and is therefore very appropriate for this task. Furthermore, in contemporary Standard Slovenian (SS) both tonal and non-tonal varieties are permitted. In Slovenian,3 the majority of central dialects, i.e., those of the Upper and Lower Carniola regions, are tonal. Additionally, Carinthian dialects in Austria and Italy are tonal, as well as the Littoral dialects of Ter, Nadiža, and Upper Soča Valley. In Rov-tarsko dialects, only Horjul and parts of Tolmin dialects are tonal. Tonal speech is found in Bela Krajina as well. Other dialects (most of the Littoral dialects, all of Styrian and Pannonian dialects, and Carinthian dialects in Slovenia) are non-tonal (cf. Rigler 1968). Srebot Rejec (1988) disputed the tonal contrast in educated speech of Ljubljana, believed to be the most important in contemporary standardization processes. She concludes: »The lexical (phonological) function of the two accents is on the wane, while the phonetic characteristics, the sing-song effect, is retained.« (Srebot Rejec 2000: 66.) Relatively recent tone loss has also been documented in Eastern Haloze (Lundberg 2003). - Slovenian has two lexical tones, acute and circumflex. For acoustic analyses of tones in Slovenian, see Vodušek 1961, Toporišič 1967, 1968, Neweklowsky 1973, and Srebot Rejec 1988, 2000. Phonetically, the acute is realized as a rising tone (or low on the stressed and high on the post-stressed/final syllable), the circumflex as the opposite. Phonologically, both tones can occur only in traditionally (i.e., diachronically) long vowels, while short vowels are considered circumflex (unmarked) in SS. In contrast to phonological limitations of better known pitch-accent languages, like Swedish and Serbo-Croatian, the contrast is preserved also in words with final stress (e.g., pot /'po:t/ - acute 'path', circumflex 'sweat'). A total of less than 100 morphologically non-related minimal pairs in tone exist (e.g., kila, kura, mula, šibica, šalica), while morphologically related pairs are abundant. In comparing the tonal and the non-tonal varieties of SS, other issues, such as inherent phonetic distinctions in vowel height, not limited to a certain prosodic feature, 2 In the present article, the term tonal (language) is used in reference to lexical tones, i.e., in this meaning of the pitch accent (as opposed to non-tonal). The term tonal is preferred to the term pitch-accented. 3 This paragraph and the corresponding references do not appear in the Slovenian version of the article. may arise. These are to be acknowledged as well, although these are not the main aim of the study. The sole nature of the linguistic material used (see section 2 for further details) renders it impossible to exclude such variables. 2 Method The present study of SS vowels is based on the extensive corpus of 241 one-, two-, and three-syllable words, compiled according to the suprasegmental criteria (stress, tone, duration).4 The list was exported to PowerPoint program and randomized manually, so that each word appeared twice non-consecutively. Speakers were instructed to read the words in citation form as they appear on the computer screen. 10 native speakers of Slovene were chosen, representative by sex (5 female and 5 male), tone contrast (5 non-tonal in origin, and 4 tonal), and age (35 years on average). The geographical criteria (i.e., the origin of the speakers) were in favour of central Slovenia. Recordings took place in the studio of the Department of Phonetics in Zagreb (Croatia) in April 2004 and in the studios of Radio Slovenia in June 2004 (1 speaker only). Sampling frequency was 44.1 kHz, at a 16-bit rate. F1-F4 of the total of 5,960 vowels were measured using Praat LPC-analysis software (ver. 4.2-4.2.14) under default settings. Typically, the individual formant steady state was measured, if possible. Alternatively, the central point or averaged value of the transient formant was measured. Altogether, 21,220 readings (of stressed and unstressed vowel formants) were acknowledged, and 4.59 % of the readings were discarded. Data were averaged and analyzed statistically (ANOVA) separately for both groups of speakers. - For a more detailed description of the speakers, method, procedures and more general results see Jurgec 2005b. 3 Results The measurements of formant frequencies were grouped into prosodic combinations (or accent types), i.e., acute, circumflex and short vowels,5 separately for both tonal and non-tonal SS. For each, mean value, standard deviation (SD), sample size, and confidence interval were calculated. One needs to note that sample size varies considerably, which is a consequence of (1) phonological distribution or constraints, (2) lexical realization, and (3) discharged cases due to nature of pronunciation. These data are presented in Table 1 -2 below. Here, F1 -F4 values are presented, while in the rest of the article only F1 and F2 are discussed. Generally, several types of differences between the tonal and the non-tonal speakers can be observed. Mean values of individual phonemes differ substantially in high-mid vowels /e/ and /o/, which have lower F1 in the tonal SS, while /e/ has somewhat higher F1. Short [a] is considerably centralized (i.e., has lower F1) for the tonal speakers, and this phenomenon is much higher than in other vowels. In /u/, the mean values of F1 are only slightly lower for the tonal speakers. 4 The complete list of words can be obtained from the author. 5 For discussion on this matter and its implications to the traditional grammar (e.g. Toporišič 2000 and the predecessors), see Jurgec 2005b: 128-131. N /e/ Id /a/ hi hi lol /u/ f1 Acute 274 357 564 731 482 587 393 304 31.92|l20| 5.71 32.9718017.23 60.91 168114.48 77.64 |l10| 14.51 41.22150 |l 1.43 67.01 |58|l7.24 41.8419018.65 54.78190111.32 Circumflex 274 373 573 725 500 608 411 304 26.57 120 4.75 41.48 |l 20 7.42 69.34 lOel 13.06 75.35 120 13.48 39.19 118 7.07 56.52 |80|l 2.36 41.49 120 7.42 42.66 119 7.67 Short 283 / 591 661 / 623 / 327 37.85 50 10.49 66.97 60116.95 97.13 50 26.92 45.10 |6o|l 1.41 47.80 20 20.95 f2 Acute 2317 2310 1969 1262 1383 1004 769 827 248.19114 45.56 244.6l| 75 55.36 311.36 68174.00 110.42110 20.63 118.48 50 32.64 83.10156|21.39 92.00 90 19.01 141.08 90 29.15 Circumflex 2293 2318 1850 1233 1350 1020 803 890 Wll 1 IIIW/\ 274.94117 49.82 235.07|l 16 42.78 291.94 lOsI 55.06 103.17120 18.46 143.6711825.92 76.49 |80|l 6.76 83.75 12014.98 156.97118 28.32 Short 2299 / 1819 1268 / 1042 / 857 271.78 48 76.89 257.95 59165.82 117.94 50 32.69 60.67 |6o|l 5.35 93.56 20 41.00 f3 Acute 2947 2839 2680 2650 2431 2689 2678 2533 355.81 120 63.66 274.39178 60.89 305.04 68 1 72.50 197.99107 37.52 206.01 50 57.10 217.0l|58|55.85 303.61 89 63.08 238.34 89 49.52 Circumflex 2916 2848 2640 2668 2554 2723 2706 2519 340.42117 61.68 261.70|l 16 47.62 329.12|l08| 62.07 194.19120 34.75 195.5811835.29 241.68|79|53.29 243.3811843.91 253.81119 45.60 Short 2858 / 2607 2581 / 2627 / 2506 335.77! 47! 95.99 225.99! 60157.18 283.17149179.29 i90.75l59l4e.67 211.68120192.77 f4 Acute 3836 3828 3884 3825 3719 3733 3591 3661 Table 1. Average values of formant frequencies (in Hz) of tonal speakers, according to phoneme, formant, andprosodic combination. Below the mean values, standard deviation, sample size, and confidence interval (± of mean value, a = .05) are listed. SD is similar in both varieties of SS, on average. Coefficient of SD is 11.22 % for the non-tonal and 10.55 % for the tonal variety, although the individual SDs for several phonemes and prosodic combinations vary. This is further discussed in section 4. On the other hand, comparison of prosodic combinations within their phonemic domain reveals fundamental differences between the two varieties of SS. Acute, circumflex and for most phonemes also short vowels are clearly much more dispersed in the tonal SS. This is clearly visible from Fig. 1, where the more dispersed accent types of the tonal SS are depicted with empty symbols (as opposed to the full symbols of the non-tonal variety). To evaluate the statistical significance of the differences among prosodic combinations a single-factor ANOVA was performed for each of the combinations. In F1, there are no statistically significant (p < .05) differences between the accent types, for all phonemes in the non-tonal variety of SS. In the tonal SS however, accent types are statistically distinct for /e/ and /o/. For /a/ the difference between long and short is highly significant (but no difference between acute and circumflex). The distinctions in /e/ and /e/ are marginal, as there is statistical significance only between 1 /i/ 1 /e/ 1 /e/ i /a/ i /s/ i hI | /o/ | /u/ f1 Acute 274 357 564 731 492 587 393 304 31.92 120 5.71 32.97 80 7.23 60.91168114.48 77.64 110! 14.51 41.22 50 11.43 67.01 5817.24 41.84 90! 8.65 54.78 90 11.32 Circumflex 274 373 573 725 500 608 411 304 26.57 120 4.75 41.48 120 7.42 69.34 |l08| 13.08 75.35 120! 13.48 39.19 118 7.07 56.52 8012.38 41.48 120! 7.42 42.66 119 7.67 Short 283 / 591 661 / 623 / 327 37.85150110.49 66.97! 60116.95 9713! 50126.92 45.10 |6o|l1.41 4780120120.95 f2 Acute 2317 2310 1969 1262 1383 1004 769 827 248.19114 45.56 244.61 75 55.36 311.36| 68174.00 110.42110! 20.63 118.48 50 32.84 83.10 5821.39 92.00 90 !l 9.01 141.08 90 29.15 Circumflex 2293 2318 1650 1233 1350 1020 803 890 274.94|l17| 48.82 235.07|l16|42.78 291.94|l08| 55.06 103.17|l20! 18.46 143.67|l18|25.92 76.49 |80|l6.76 83.75 |I2O!I4.98 156.97|l18| 28.32 Short 2299 / 1819 1268 / 1042 / 857 271.78 48 76.89 25795! 59! 65.82 117.94 50132.69 60.67 6015.35 93.56 20 41.00 f3 Acute 2947 2839 2680 2650 2431 2689 2678 2533 355.8l|l20| 63.86 274.39|78|60.89 3O5.O4168172.50 19799|l07!3752 2O6.O1150157.IO 217.01158|55.85 303.61189 !63.08 238.34| 89148.52 Circumflex 2916 2848 2640 2668 2554 2723 2706 2519 340.42117 61.68 261.7011647.62 329.12|l08| 62.07 19419120! 34.75 195.5811835.29 241.687953.29 243.38 118!43.91 253.81119 45.60 Short 2858 / 2607 2581 / 2627 / 2506 335.77 47 95.99 225.99|60|5718 283.17 49179.29 190.755948.67 211.68 20 92.77 f4 Acute 3836 3828 3S84 3825 3719 3733 3591 3661 380.30111 70.75 363.90 78 80.76 450.3?! 66 |l 08.65 371.66108! 70.09 348.04 50 96.47 309.985879.78 379.54 88 !79.30 407.23 88 65.08 Circumflex 3846 3848 3678 3853 3703 3772 3617 3629 424.67114 77.96 412.2711575.35 437.27!l 06! 83.24 351.46119! 63.15 329.4611859.44 278.427961.40 363.69 119!65.34 429.93117 77.90 Short 3796 / 3799 3763 / 3692 / 3573 397.22 49 111.22 411.82|60|l04.20 353.38 47 |l 01.03 298.326075.48 468.60 20 205.37 Table 2. Average values of formant frequencies (in Hz) of non-tonal speakers, according to phoneme, formant andprosodic combination. Below the mean values, standard deviation, sample size and confidence interval (± of mean value, a = .05) are listed. most distinct prosodic combinations, i.e., acute and short (but not between acute and circumflex, and circumflex and short). In F2, statistical significance is attested for both accent types of /o/ in the tonal SS. Acute and circumflex difference is significant also in [e], [a], [u], circumflex vs. short in [a], and acute vs. short in [e] and [e]. In [a], significance is only marginal. In sum, the accent types of [a] and of both tense mid vowels [e], [o] differ significantly, while in [e] and [e] this effect is only marginally significant. There is no statistical significance only among the accent types of the high vowel [i] and central vowel [a]. Detailed results of the analysis for both F1 and F2 are presented in Table 3. 472 General Linguistic Topics 2250 2000 1750 ^"^^1500 1250 1000 /i/ /u/ ▲ ■ /e/ /@/ ▲ ■ ' J ' I Table 3. F1xF2 v^^wel space of tonal and non-tonal varieties of SS. This is not the case in non-tonal SS, where no variability is attested in F1. In F2 however, a marginal statistical significance is found in [e], [a] and [u] (see Table 4 for further results). This fact is explained in Section 4. 4 Discussion and conclusion Previous section revealed several differences between the groups of tonal and non-tonal speakers, either related to purely acoustic phonetic factors of tone itself or not. As regards the latter, one could say that in the tonal variety, low-mid and high-mid vowels are less central. [e] and [o] are therefore more tense perceptually, or higher articulatorily in the tonal SS than in the non-tonal, while [e] is lower. The only exception is [e], which exhibits no such tendency. Generally, in Slovenian spoken in central Phoneme Fl F2 Accent types df F p(a=.05) Accent types df F p (a=.05) N Acute vs. circumflex 1,238 .005 .942 Acute vs. circumflex 1,229 .497 .481 Acute vs. short 1,168 2.30 .131 Acute vs. short 1,160 .156 .694 Circumflex vs. short 1,168 2.68 .103 Circumflex vs. short 1,163 .022 .882 lei Acute vs. circumflex 1,198 8.43 .004 Acute vs. circumflex 1,189 .055 .814 Id Acute vs. circumflex 1,174 .852 .357 Acute vs. circumflex 1,174 6.67 m Acute vs. short 1,126 5.64 .019 Acute vs. short 1,125 8.65 .004 Circumflex vs. short 1,166 2.48 .177 Circumflex vs. short 1,165 .461 .498 la! Acute vs. circumflex 1,228 .400 .528 Acute vs. circumflex 1,228 4.21 :Q41 Acute vs. short 1,158 23.71 <.00001 Acute vs. short 1,158 .118 .731 Circumflex vs. short 1,168 21.01 <.00001 Circumflex vs. short 1,168 3.84 .052 Id Acute vs. circumflex 1,166 1.41 .237 Acute vs. circumflex 1,166 2.04 .155 hi Acute vs. circumflex 1,136 4.04 .047 Acute vs. circumflex 1,136 1.40 .239 Acute vs. short 1,116 12.16 .0007 Acute vs. short 1,116 8.41 .004 Circumflex vs. short 1,138 3.01 .085 Circumflex vs. short 1,138 3.54 .062 lol Acute vs. circumflex 1,208 9.29 .003 Acute vs. circumflex 1,208 7.86 .006 /U/ Acute vs. circumflex 1,207 .000 .992 Acute vs. circumflex 1,206 9.07 .003 Acute vs. short 1,108 2.87 .093 Acute vs. short 1,108 .791 .376 Circumflex vs. short 1,137 4.55 .035 Circumflex vs. short 1,136 .876 .351 Table 4. Single-factor ANOVA results for separate phonemes and prosodic combinations of the tonal SS. The default Alpha factor is used (.05). Statistically significant values are underlined; marginally significantp-values (0.035-0.055) are marked with a dashed line. dialects, including Ljubljana, the feature [+ ATR] has greater effect on vowel quality, decreasing F1 of high-mid vowels. This is complemented by the increased F1 of low-mid, but the effect is rather limited. The above-mentioned phonetic property is consistent with experimental data from non-central Slovenian in Ozbič 1998ab for SS as spoken in Trst (Trieste) and in Jurgec 2005a, for speech of Ovčja vas (Valbruna). One should also take into account the gender of both groups of speakers: 3 females and 2 males are tonal (the situation is reversed for the non-tonal speakers). Average F0 of females is higher than that of males, and there is a positive correlation between average F0 and formant frequencies. Therefore, the increased F2 of tonal speakers in /e/, /e/, and /i/ can be attributed to this, but no such effect should be present in F1. Moreover, certain phonological variables influence formant frequencies of the tonal variety. Quantity contrast in SS stressed vowels is at least questionable (Sre-bot Rejec 1988, Petek et al. 1996), if not already completely neutralized, at least for speakers of Ljubljana, as well as for most speakers in southwest and northeast Slovenia. On the other hand, these oppositions are still present on dialectal level and Phoneme F1 F2 Accent types df F p(a=.05) Accent types df F p(a=.05) N Acute vs. circumflex 1,238 1.81 .179 Acute vs. circumflex 1,237 .279 .598 Acute vs. short 1,168 .371 .543 Acute vs. short 1,168 .695 .406 Circumflex vs. short 1,168 .211 .647 Circumflex vs. short 1,167 .220 .640 lei Acute vs. circumflex 1,195 .141 .708 Acute vs. circumflex 1,194 .189 .665 lei Acute vs. circumflex 1,174 .483 .488 Acute vs. circumflex 1,173 6.59 .011 Acute vs. short 1,116 .012 .914 Acute vs. short 1,116 6.11 .015 Circumflex vs. short 1,154 .275 .600 Circumflex vs. short 1,153 .043 .836 lal Acute vs. circumflex 1,228 .009 .924 Acute vs. circumflex 1,228 7.45 .007 Acute vs. short 1,158 .321 .572 Acute vs. short 1,158 .018 .893 Circumflex vs. short 1,168 .308 .580 Circumflex vs. short 1,168 4.77 .030 Isl Acute vs. circumflex 1,159 .309 .579 Acute vs. circumflex 1,159 .372 .543 hi Acute vs. circumflex 1,113 .340 .561 Acute vs. circumflex 1,113 .0006 .980 Acute vs. short 1,91 .559 .456 Acute vs. short 1,91 .671 .415 Circumflex vs. short 1,108 1.72 .192 Circumflex vs. short 1,108 .670 .415 lol Acute vs. circumflex 1,206 .216 .643 Acute vs. circumflex 1,206 1.07 .303 /U/ Acute vs. circumflex 1,208 1.06 .304 Acute vs. circumflex 1,207 7.60 .006 Acute vs. short 1,108 3.46 .066 Acute vs. short 1,107 .0005 .994 Circumflex vs. short 1,138 1.78 .184 Circumflex vs. short 1,136 2.20 .140 Table 5. Single-factor ANOVA results for separate phonemes and prosodic combinations of the non-tonal SS. The default Alpha factor is used (.05). Statistically significant values are underlined; marginally significant p-values (0.035-0.055) are marked with a dashed line. in the sub-standard speech as qualitative changes, i.e., phonological reduction processes. Thus when speaking SS, speakers tend to avoid these processes, and since they are unable to produce any quantity contrasts, diachronically short vowels merge with unreduced long vowels (Rigler 1968). Present data confirm only marginally significant contrast between short and long vowels, limited to the tonal SS, namely to the phonemes /e/ and /e/, in F1 and F2 (see Table 3-4). The only exception is /a/, where phonologically short [a] is considerably centralized. The average F1 of short [a] is 67 Hz lower than the average F1 in long [a]. This is highly significant (p < 0.0001), although the coefficient of SD is moderately increased (14.7 % in F1). This unique phenomenon, not attested in other phonemes, can be corroborated by the data in Petek et al. 1996, where /a/ was the only phoneme that exhibited (some) durational differences. This inconsistency has not been explored yet and has had no influence on normative practice so far. As regards the influence of phonological tone on formant frequencies, the results prove a positive correspondence. To confirm the research hypothesis, one should first prove that there are differences in formant frequencies of the tonal SS and that they are statistically significant. Furthermore, that no such differences exist in the non-tonal SS, and that this situation cannot be explained otherwise, for example as a consequence of other phonetic features. Suprasegmental (phonological) variables are statistically significant in majority of phonemes in the tonal SS (Fig. 1). Upon further inspection (ANOVA, cf. Table 3-4), only /i/ and /s/ exhibit no significant differences between the accent types. /s/ is a phonetically neutral vowel and attested differences should not be contraindica-tive to the research hypothesis. On the other hand, the same situation in /i/ cannot be explained in terms of general phonetics. However, other data from Slovenian and its formant frequencies (Jurgec 2005bc), posit an interesting property of Slovenian /i/, being the least subjected to influences of stress and word-position. In contrast, another high vowel, /u/, is subjected to much greater degree of variance, while the influence of tone is only marginal. In the non-tonal SS, individual accent types of each phoneme are clearly less dispersed. This is evident from Fig. 1 (e.g., phonemes /e/, /o/, /e/, and /a/), and corroborated by statistical analyses in Tables 3-4. In F1, no prosodic differences are statistically significant. In F2 however, there are a few exceptions: acute [e] is distinctive of circumflex and short, as it is circumflex [a]. There is also statistical significance in acute or circumflex [u]. Dispersion in [e] could be attributed to the problematic distribution of both front mid vowels, which are morphonologically connected, and the distribution in SS differs greatly from the contemporary dialectal and sub-standard realization. When unstressed, both phonemes are neutralized and merged into a single archiphoneme (Le-histe 1961, Srebot Rejec 1988, 1998), which is realized as [e] in the pre-stressed and as [e] in the post-stressed position (see Jurgec 2006 for further data and discussion). This is corroborated by the increased coefficient of SD, which is exhibited in both front mid vowels of the non-tonal SS; in F1 of [e] the coeff. is 20.1 %, almost twice the average, in [e] it is 15.3 % (F2 of both vowels is too close to influence SD). Although erroneous cases of pronunciation were discharged (see the drop in sample size of both phonemes in Table 2), a partial overlap in formant frequencies is a possible and also probable explanation. The increase is also noticeable in back mid vowels (yet lower than in front vowels) and in [e] of the tonal SS (but not in front vowels and [e]) and exhibits a general phonological tendency of contemporary Slovenian. To sum up, the data of the non-tonal [e] should be regarded highly inconclusive. The increased coefficient of SD is observed in [u] as well, both tonal and non-tonal (on average, well above 15 % in F2). The fact that circumflex [u] is statistically distinct from the acute and short is also surprising. In most vowels, circumflex is more similar to short than the acute, which is in accordance with the traditional theory that considers phonologically short vowels circumflex in tone. As the significance is similar in both varieties of SS, one can say that the analysis is dubious: [u] must also be influenced by other variables. For example, the difference between word-final and initial vs. medial position of the two high vowels, documented in certain sources (e.g., Toporišič 2000: 50). The present analysis, based on linguistic material of the existing and generally known words in Slovene, cannot answer this problem satisfactorily. This will be done in the future work. The phoneme /a/ has a moderately increased coefficient of SD as well, under acute tone more than under circumflex and as short. One reason for this could be a considerable backness of the low vowel in Styrian and Pannonian dialects, where three of our speakers originate.6 If this is true, only the acute is being influenced and is statistically significantly differs from circumflex and short [a]. This cannot be caused by the phonetic factors per se, but by dialectal phonetic influences and should therefore be disregarded. All things considered, vowel formant frequencies of the tonal SS are affected by phonological tone. The differences may not be large (as opposed to influence of consonantal environment, stress, and certain extralinguistic factors), but they are still significant, and, as a rule, not present in the non-tonal speech. Whether this is directly related to the distinctions in fundamental frequency or intensity attested in Slovene acute vs. circumflex tone, remains unknown. However, F0 and formant frequencies show a positive correlation (via stress, gender or speaking style), and the correspondence grows exponentially, higher formants exhibiting much larger increase than the lower ones if F0 rises. Intensity (via duration, stress or speaking style) also corresponds to formant frequencies, i.e., vowels with greater intensity have higher formant frequencies (either via duration, stress, or speaking style), all other things being equal. - The design of the present experiment itself renders it impossible to account for all acoustic and articulatory factors and to determine their extent. It proves, however, that such differences occur. References Peter F. Assmann and Terrance M. Nearey, 1987: Perception of front vowels: The role of harmonics in the first formant region. The journal of the Acoustical society of America LXXXI/2. 520-534. Juraj Bakran, 1989: Djelovanje naglasaka i dužine na frekvencije formanata vokala. Govor VI/2. 1-12. D. G. Childers and Ke Wu, 1991: Gender recognition from speech. Part II: Fine analysis. The journal of theAacoustical society of America XC/4. 1841-1856. Olle Engestrand, 1988: Articulatory correlates of stress and speaking rate in Swedish VCV utterances. The journal of the Acoustical society of America LXXXIII/5. 1863-1875. Donna Erickson, 2002: Articulation of extreme formant patterns for emphasized vowels. Pho- netica IL/2-3. 134-149. Marios Fourakis, 1991: Tempo, stress, and vowel reduction in American English. The journal of the Acoustical society of America XC/4,1. 1816-1827. Marios Fourakis, Antonis Botinis in Maria Katsaiti, 1999: Acoustic characteristics of Greek vowels. Phonetica LVI/1-2. 28-43. Thomas Gay, 1978: Effect of speaking rate on vowel formant movements. The journal of the Acoustical society of America LXIII/1. 223-230. 6 There were no cases of non-standard rounded back vowel [d], which differs from SS low vowel considerably, and would subsequently be excluded from further analysis. Peter Jurgec, 2005a: Fonetični opis govora Ovčje vasi. Ovčja vas in njena slovenska govorica / Valbruna e la sua parlata slovena. Kanalska dolina: Slovensko kulturno središče Planika, ZRC SAZU. 60-84. — 2005b: Formant frequencies of standard Slovenian vowels. Govor XXII/2. 127-144. — 2005c: Položaj v besedi in formantne frekvence samoglasnikov (standardne slovenščine), I. Naglašeni samoglasniki. Jezikoslovni zapiski XI: 1. 121-132. — 2006: O nenaglašenih /e/ in /o/ v standardni slovenščini. Slavistična revija LIV/2. 173-185. Ilse Lehiste, 1961: The phonemes of Slovene. International journal of Slavic linguistics and poetics IV. 48-66. Ilse Lehiste and Pavle Ivic, 1963: Accent in Serbo-Croatian: An experimental study. Ann Arbor: University of Michigan (Michigan Slavic Materials 4). Björn Lindblom, 1963: Spectrographic study of vowel reduction. The journal of the Acoustical society of America XXXV/11. 1773-1781. Grant H. Lundberg, 2003: Typology of tone loss in Haloze, Slovenia: An acoustic and autoseg-mental analysis. Slovenski jezik / Slovene linguistic studies III. 169-189. James D. Miller, 1989: Auditory-perceptual interpretation of the vowel. The journal of the Acoustical society of America LXXXV/5. 2114-2134. Seung-Jae Moon and Björn Lindblom, 1994: Interaction between duration, context, and speaking style in English stressed vowels. The journal of the Acoustical society of America XCVI/1. 40-55. Thomas Murry and Sadanand Singh, 1980: Multidimensional analysis of male and female voices. The journal of the Acoustical society of America LXVIII/5. 1294-1300. Gerhard Neweklowsky, 1973: Slowenische Akzentstudien [...]. Vienna: Verlag der Österreichischen Akademie der Wissenschaften. Martina Ozbič, 1998a: Akustična spektralna FFT-analiza samoglasniškega sistema slovenskega jezika: formanti slovenskih samoglasnikov. Jezikovne tehnologije za slovenski jezik: Zbornik konference. 55-59. Http://nl.ijs.si/isjt98/zbornik/sdjt98-Ozbic.pdf. — 1998b: Razmerja med formanti samoglasnikov matične in tržaške slovenščine. Uporabno jezikoslovje VI: Jezikovne tehnologije. 124-135. Bojan Petek, Rastislav Šuštaršič and Smiljana Komar, 1996: An acoustic analysis of contemporary vowels of the Standard Slovenian language. Proceedings ICSLP 96: Fourth International Conference on Spoken Language Processing, October 3-6, 1996, Philadelphia, PA, USA. 133-136. Http://www.asel.udel.edu/icslp/cdrom/vol1/820/a820.pdf. Michel PiTERMANN, 2000: Effect of speaking rate and contrastive stress on formant dynamics and vowel perception. The journal of The acoustical society of America CVII/6. 3425-3437. Jakob Rigler, 1968: Problematika naglaševanja v slovenskem knjižnem jeziku. Jezik in slovstvo XIII/6. 192-199. Adrian P. Simpson, 2001: Dynamic consequences of differences in male and female vocal tract dimensions. The journal of the Acoustical society of America CIX/5.1. 2153-2164. Agaath M. C. Sluijter and Vincent J. van Heuven, 1996: Spectral balance as a acoustic correlate of linguistic stress. The journal of the Acoustical society of America C/4.1. 2471-2485. Tatjana Srebot Rejec, 1988: Word accent and vowel duration in Standard Slovene: An acoustic and linguistic investigation. Munich: Otto Sagner (Slavistische Beiträge, 226). —1998: O slovenskih samoglasniških sestavih zadnjih 45 let. Slavistična revija XLVI/4: 339-346. — 2000: Ali je današnja knjižna slovenščina še tonematična? Razprave II. razreda SAZU XVII. 51-66. Jože Toporišič, 1967: Pojmovanje tonemičnosti slovenskega jezika. Slavistična revija XV/1-2. 64-108. —1968: Liki slovenskih tonemov. Slavistična revija XVI. 315-393. — 2000: Slovenska slovnica. Maribor: Obzorja. Betty Tuller, Katharine S. Harris and J. A. Scott Kelso, 1982: Stress and rate: Differential transformations of articulation. The journal of the Acoustical society of America LXXI/6. 1534-1543. R. J. J. H. Van Son and Louis C. W. Pols, 1992: Formant movements of Dutch vowels in text, read at normal and fast rate. The journal of the Acoustical society of America XCII/1. 121-127. Božo VODUŠEK, 1961: Grudsätzliche Betrachtungen über den melodischen Verlauf der Wortakzente in den zentralen Slowenichen Mundarten, Linguistica IV. 20-38. Ke Wu and D. G. Childers, 1991: Gender recognition from speech. Part I: Coarse analysis. The journal of the Acoustical society of America XC/4. 1828-1840. Povzetek Članek predstavlja formantne frekvence samoglasnikov tonemske in netonemske različice standardne slovenščine (SS). Upoštevajoč fonološko distribucijo in nadsegmentne lastnosti je bil sestavljen obsežen korpus eno-, dvo- in trizložnic. 241 besed je v naključnem vrstnem redu izolirano bralo deset govorcev, enakomerno porazdeljenih po spolu, izvoru in tonemskosti. Pet jih je bilo tonemskih (3 ženske in 2 moška govorca), pet netonemskih. Snemanje je bilo digitalno, pri standardnih pogojih, tj. frekvenci vzorčenja 44,1 kHz in 16-bitni kvantizaciji. F1-F4 skupno 5.960 samoglasnikov so bili izmerjeni z LPC-analizo v programu Praat, pri standardnih nastavitvah. Izmerjene vrednosti so bile razvrščene v skupine in izračunana povprečja. Sledila je statistična analiza, vključno z analizo variance (ANOVA). Za podrobnejše podatke gl. Jurgec 2005b. Povprečne vrednosti (skupaj s standardnim odklonom, številom meritev in intervalom zaupanja) obeh različic SS so v prikazih 1 in 2 (Table 1 in 2). V prikazu 3 (Figure 3) je akustični diagram F2/F1 za tonemske (prazni simboli) in netonemske (polni simboli) govorce. V prikazih 4 in 5 (Table 3 in 4) pa so rezultati analize variance (najprej za tonemske, potem za netonemske govorce). Rezultate lahko razdelimo v dve skupini, ki so bodisi (nad)narečni v fonetičnem in fonološkem smislu ali strogo akustični. V prvi skupini so tako razlike v F1 [+ ATR] srednjih samoglasnikov [e] in [o], ki je v tonemski slovenščini nižji, kratki [a] je pri tonemskih govorcih občutno centraliziran, česar ni pri drugih samoglasnikih tonemskih ali netonemskih govorcev. Razlike med akutiranimi, cirkumflektiranimi in kratkimi samoglasniki posameznega fone-ma so pričakovano večje v tonemski SS in povečini tudi statistično značilne v F1 in/ali F2. Ni pa take razlike pri [s] in [i]. Pri [u] so očitno pomembnejše druge segmentne spremenljivke, saj se F1 in F2 obeh skupin tu bistveno ne razlikujejo. Sicer so v netonemski SS statistično značilne razlike redke; tako lahko F2 pri [a] pojasnimo z narečnimi vplivi, pri [e] pa je problematična distribucija. Razpršenost skupin akutiranih, cirkumflektiranih in kratkih samoglasnikov ter njihovih formantnih frekvenc v tonemski SS lahko razložimo (tudi) kot posledico osnovne frekvence in jakosti na eni ter jakosti in fonetične redukcije oz. učinka podhranjenosti (undershoot) na drugi.