Acoustic analysis of Croatian and Serbian RP pronunciation – formant analysis and fundamental frequency measurements

Bašić, Iva. (2018). Acoustic analysis of Croatian and Serbian RP pronunciation – formant analysis and fundamental frequency measurements. PhD Thesis. Filozofski fakultet u Zagrebu, Department of Linguistics.
(Poslijediplomski doktorski studij lingvistike) [mentor Varošanec-Škarić, Gordana].

Preview

PDF (Croatian) Download (5MB) | Preview

Abstract

The subject of this study is on the one hand motivated by the need for the necessary acoustic description of the vowel system of Croatian, and on the other hand by the recent research methods in forensic phonetics. Formant analysis is the main component of most forensic phonetic cases. Physiological features of the speaker, their sociolinguistic background and idiosyncratic phonetic behaviour, are reflected in formant frequencies. That is the reason why formant analysis is applied in different scientific fields: articulatory, acoustic and forensic phonetics, sociophonetics, sociolinguistics, etc. The primary aim of this research is to determine the reference formant frequencies (F1-F3) of the vowel system of the Croatian and Serbian languages. For the purposes of this research, 184 native speakers were recorded. After the verification process, 162 speakers were selected from the corpora. Both languages were represented by an equal number of speakers (NCRO= 81 and NSER= 81), with a remotely larger number of male compared to female voices (NF= 70 and NM= 92). Ladefoged (2003) recommended that corpora in sociolinguistic and sociophonetic research should include speakers of both sexes, due to the differences in coexistent phonology, phonetics etc., which is confirmed in the majority of world languages. Speakers recorded for this dissertation were chosen according to five criteria: speech status, place of birth and an extended stay, place of birth of their parents, level of education, and birth year. The recordings were carried out in very similar conditions: in rooms with reduced noise level or in studio conditions. All speakers were recorded by the same recording schedule, and were given the same instructions. Vermeulen and Cambier-Langveld (2017) noted that the same speech style (reading, spontaneous speech, etc.) is optimal for speaker comparison in forensic phonetics. For the purposes of this research, speakers were instructed to read a list of 50 shorter sentences in which target words were placed in the final positions. Each vowel was represented through 10 two-syllable words with a different phonetic environment. Formant frequencies (F1-F3) were estimated from the central stable part of the accented vowel of the target word, with the help of the Praat program (Boersma & Weenik, 2015). In some vowels ([u], [o] and [i]), and more often within female voices, formants tended to overlap. This spectral integration was noticed in both languages, and in these samples all results were subsequently acoustically and perceptively checked and corrected. The fundamental frequency was also evaluated in the central part of the accented vowel. Very low frequencies, which were the result of the glottal fry, have been excluded from the results. Glottalization was more often recorded within men voices, which is in sociophonetics interpreted as a possible social marker of highlighting their own masculinity. The programs MATLAB (MathWorks Inc., 2015) and JASP (JASP Team, 2018) were used for the statistical analysis of the collected data. Descriptive statistics consisted of determining average, median, minimum, and maximum values of formants and F0. For the purposes of this research, the frequency ranges of F0 were also calculated. They were determined by specifying the average minimum and maximum values of the fundamental frequencies for each speaker individually. The results were then processed by factors of different sex and language. Further analyses were conducted using frequency values of formants (F1-F3) and the fundamental frequencies. For the purpose of testing the significant differences in average absolute deviation of these, acoustic parameters (as dependent variables) have been calculated. The average absolute deviation has been selected as a more stable measure for the dispersion of measured results (compared to the standard deviation, or variance analysis), given the increasing number of measurements. In this way, the measured dispersions of results have been compared amongst different vowels, sexes, and languages. Different parametric tests have been used for the comparison of differences between the various groups of speakers (of diverse sex and/or language). Correlation coefficients have been calculated between formant frequency values and the fundamental frequency, and also between formants themselves. Correlation coefficients have been calculated using Pearson formulas for estimating connectedness. The study examined sex and language differences among analysed speakers in several acoustic parameters (formant values, formant dispersion, and fundamental frequency). Furthermore, it questioned if there are some acoustic differences in the variability of vowel systems according to the factors of sex and language. Since it is generally known that coarticulation has the strongest impact in the trajectory areas of vowels, this study questioned whether a different phonetic environment has an impact on the formant values in the central part of the accented vocals. Given that some authors emphasize that it is more useful to interpret the relations among formants than the average values for each formant separately (Chistovich & Lublinskaya, 1979; Chistovich, 1985; Hayward, 2000; Harrington, 2013), in this research differences in formant relations between speakers of different sexes were analysed, as well as differences among speakers of different languages. This study presumed that the majority of male speakers would have lower frequency values of the formants, fundamental frequency, and measure of formant dispersion, compared to those of female speakers. Considering the fact that some studies showed that coarticulation influence is the strongest around vowel [a] due to its lowest articulation stability (Stevens & House, 1963), and that in sociophonetical research of Croatian (Škarić, 2009; Kišiček, 2012) the same vowel was described as the most distinctive vowel in Croatian, the investigation was also directed towards the variability of formant frequencies in different vowels. Since the pilot research of Varošanec-Škarić, Bašić and Kišiček (2016) has shown that vowel [a] was more open, vowels [i] and [e] were more front, vowels [i] and [u] more closed, and vowel [u] more back in Croatian than in Serbian, in the present study the vowel systems of the analysed languages have been acoustically described and compared. Furthermore, it was expected that overall average values of the fundamental frequency would be lower for Croatian speakers of both gender groups compared to the results in previous studies (Škarić, 1998; Jovičić, 1999; Biočina, Varošanec-Škarić & Kišiček, 2017). Also, this study determined the frequency ranges of the fundamental frequency for speakers of both languages, as well as for both gender groups. Correlations of formants and the fundamental frequency, as well as correlations between the formants themselves have been examined. Results of the formant analysis have shown that both genders in both languages have the lowest average values of the first formant while pronouncing the front vowel [i], and the highest F1 values for the central vowel [a]. The lowest average values of the second formant have been found for the back vowel [u] and the highest for [i]. The third formant had the highest average values for vowel [i], whereas [o] had the lowest. Considering that previous research of similar subject matter in the Croatian language and this research was conducted with substantial methodological differences (in the number of speakers, pooling results of speakers of different sexes, deficient speech material, different speaking style, etc.), average values of formants were compared descriptively – without statistics, which in that case would be unjustified. Results of the average reference values for Croatian were closest to the results from the study by Varošanec-Škarić and Bašić (2015), and for Serbian; they were closest to the results from Marković and Bjelaković (2009), as well as from Varošanec-Škarić et al. (2016). An overview of recent studies on formant analysis in the Croatian and Serbian languages has shown that there is a greater discrepancy in the formant values for Serbian – as stated by different authors - than it is for Croatian. In addition, this study includes an acoustic contrastive analysis in order to describe the vowel systems of the Croatian and Serbian languages. The importance of the differences was also statistically analysed. The results have shown that vowel [a] is placed further back in both genders for Croatian speakers, where a significant difference was found for the first and third formant. The front vowel [e] has shown itself as more front and it was observed that it is pronounced with less open lips than in Serbian (statically significant only among women). Between female speakers of the Croatian and Serbian languages there is also a difference in the pronunciation of the vowel [e], which is somewhat more closed in Croatian (yet it has no statistical significance). Vowel [i] is also more closed in Croatian (statistical significance determined only among men), while in terms of the feature front/back, one can say that the results are sexually dimorphic: male speakers tend to have higher values for the second formant which indicates more front pronunciation, while female speakers, on the other hand, pronounce it more to the back when compared to Serbian speakers. According to the results from the conducted analyses, back vowels [o] and [u] are more closed (and back) in Croatian, and the vowel [o] is significantly more closed only in women. Statistical significance for vowel [u] has been observed among all formants for both sexes, which brings us to the conclusion that [u] is more closed and further back in Croatian and that it is articulated with more rounded lips. After determining the average formant values in both analysed languages, the next step was to compare the variability of different formant frequencies between all vowels within both genders and languages separately. It was observed that F1 has the lowest value variability across the vowels, while it is somewhat higher for F2, and the highest for F3. If we look at the results across different vowels, the highest variability of the first formant was observed in vowel [a] in both gender groups. This was indicated on the one hand by the results from sociophonetic studies in Croatian (Škarić, 1991; Varošanec-Škarić, 2010), according to which the vowel [a] is the most distinctive vowel of the Croatian vowel system, and on the other hand by the results from studies according to which the coarticulation influence is strongest for the vowel [a], due to its lowest articulatory stability (Stevens & House, 1963). In the Croatian language, the biggest dispersion of the second formant was determined among men during the pronunciation of vowel [o], and among women during the pronunciation of the front vowel [i]. The third formant varies the most in the back vowel [u] in male speakers, and in [i] for women. In the Serbian language, the greatest dispersion of F1 in women was observed for the front vowel [e], while the back vowels [o] and [u] showed the same values of dispersion among the male population. The second formant had the highest variability in both sexes of Serbian speakers during the pronunciation of the front vowel [e], while for F3 the same was observed in the pronunciation of the back vowel [o]. Based on the comparison of average formant frequencies between different sexes in one language, and between speakers of the same sexes in both languages, several conclusions have been drawn. As expected, in both analysed languages female speakers had higher values of all analysed formants (F1-F3), compared to male speakers. By means of statistical analysis it has been confirmed that Croatian speakers of both sexes differ significantly in their average values: F1 is significant for all vowels, F2 for the majority (with the exception of central [a] and back vowel [u]), F3 also for the majority (except the back vowel [u]), as well as in values of formant dispersion (Df) in all vowels. Hence, it can be said that F1 and Df are stronger acoustic parameters for sex differentiation in Croatian than parameters F2 and F3, which has already been ascertained by Torre III and Barlow (2009). The results of the analysis in the Serbian language have also confirmed this pattern of gender distinction based on the values of F1. Namely, the results show that men and women differ with a statistical significance in their average values: F1 in nearly all vowels (with the exception of the back vowel [u]), F2 for fewer vowels (in [e] and [u]), and F3 for back vowels. Unexpectedly, the parameter of formant dispersion has been a very weak indicator of gender distinction in Serbian (no statistically significant difference has been found). Therefore, we can conclude that in the Serbian language the strongest factor for gender distinction is the first formant, while the second and third are equally weak indicators of gender differences. The measure of formant dispersion (Df) was used to examine gender and language differences between the analysed speakers. The results show that the Df values in the Croatian language are primarily higher among women, except for vowels [a] and [o]. Female Serbian speakers had higher Df values in all vowels, compared to female speakers of Croatian. The statistical significance of variability of the analysed parameters (F1, F2, F3 and Df) was examined between male and female Croatian speakers and, subsequently, between speakers of different sex in the Serbian language. The results show that the difference in dispersion of formant values in Croatian is statistically significantly higher in women for vowels [a], [e] and [i], while in men this is the case for back vowels. Therefore, we can say that acoustic dispersion of formant frequencies is higher for female speakers of the Croatian language, which reinforces equivalent results of studies in other languages (Gordon & Heath, 1998; Hanson & Chuang, 1999). For Serbian speakers, the variability of the first formant is significantly higher among female speakers in nearly all vowels (except [u]), whereas the variability of F2 was primarily higher among male speakers (in vowels [a], [i] and [o]). Apart from examining sex differences, the aim of using the measure of formant dispersion was to analyse language differences between the speakers. The results have shown that Df values are mainly higher for speakers of the Serbian language (vowels [a], [e] i [u]). The results showed that Df values are mainly higher within speakers of Serbian (for vowels [a], [e] and [u]). Higher Df values have been determined among speakers of Croatian in the front vowel [i], while in the back vowel [o] their values were very close. Female speakers of Croatian had lower Df values in all vowels, compared to female speakers of Serbian. Since the same tendency has been confirmed in both groups of speakers, these findings suggest that differences in Df values are caused by language differences, respectively by differences in the vowel systems of the analysed languages, which had reflected on formant values, as well as on Df values. In this research, the results also showed lower variability of formant frequencies (F1-F3) among speakers of both sexes in Croatian, compared to speakers of Serbian (statistically significant for vowels [e] and [o] for male speakers, and for vowels [e], [i], [o] and [u] between female speakers of Croatian and Serbian). Taking into consideration that the phonetic environment effects not only the trajectory part of the vowel, but also the formant frequencies in the stable part of the vowel, this research questioned coarticulation effects of different phonetic environments. The results showed that F1 values tend to fall in fricative and plosive phonetic environment. On the other hand, F2 values tend to rise in the same environment. An increase in F2 values is especially emphasized in the plosive environment in front vowels. Second formant showed higher values in the fricative environment in back vowels, which was confirmed in different languages (Stevens & House, 1963). The second aim of this dissertation was to compare the different measures of the fundamental frequency, with the purpose of questioning sex and language differences between the Croatian and Serbian languages. The results of the acoustic analysis and statistical data processing showed that the average F0 value for male speakers of Croatian is 118 Hz, and for female speakers 197 Hz. The highest average F0 values were calculated for the front vowel [i], and the lowest for the central vowel [a], which was confirmed for both sexes. In comparison to previous research, frequency values are very close, closest to the results of the most recent studies with similar methodology (Varošanec-Škarić, 2010; Kišiček, 2012; Biočina et al., 2016; Varošanec-Škarić et al., 2017). In the group of Serbian speakers, the average F0 for male speakers is 108 Hz, and 179 Hz for females. Comparing the fundamental frequency values between speakers of the same sex and different language has shown that speakers of Croatian (females and males) have significantly lower F0 values for every analysed vowel and generally at the level of all vowels (p<0,001). Sex and language differences have also been analysed according to the range of the fundamental frequency. Descriptive statistics showed that male speakers of Croatian mainly have wider F0 range, with regard to women. Surprisingly, in Serbian the results were opposite. Comparing the frequency ranges of F0 between the speakers of Croatian and Serbian, results showed that male speakers of the Croatian language have a wider F0 range in vowels [a], [e] and [i]. In the group of male speakers of the Serbian language, results showed a wider range for back vowels ([u] and [o]). Female speakers of Croatian and Serbian also differed in F0 range. Speakers of Serbian have shown a wider range in vowels [a], [e] and [u], while in the remaining vowels female speakers of Croatian had a wider range. Although frequency ranges are a frequently used parameter in phonetics, statistically speaking they are not a stable and reliable indicator of dispersion. Accordingly, significance of sex and language differences in F0 were tested with complex ANOVA and multiple paired t-tests. Descriptive statistics, ANOVA analysis, and t-tests suggest very diverse results. Namely, for all vowels it has been established that F0 is significantly more variable within the group of female speakers, than within male speakers. These results have been confirmed for both analysed languages, and were found in numerous sociolinguistic and sociophonetic studies for different languages. The findings of this study also indicate that there are no significant differences in F0 variability (except for vowel [e]) between speakers of the analysed languages (confirmed for both sexes). Finally, this dissertation has questioned the correlations between the fundamental frequency and formants, as well as the correlations between formants themselves. These correlations were analysed within speakers of different languages and different sexes. Also, the significance of the correlations themselves was analysed. In the group of Croatian speakers, results showed that there are statistically significant correlations between F0 and F2, as well as between F0 i F3. Surprisingly, correlations between F0 and the first formant have not been found. Taken together, these results suggest that there is a greater number of correlations between F0 and formants within the group of female speakers (for vowels [e], [i] and [o] between F0 and F2, as well as for [u] between F0 and F3), than within the group of male speakers (for vowel [i] between F0 and F2, as well as for [u] between F0 and F3). For both sexes in the group of Croatian speakers, the results showed a greater number of correlations between formants, than between formants and the fundamental frequency (in the group of male speakers for vowels [a] and [u] between F1 and F2, for [e] between F2 and F3, and in the group of female speakers for vowels [a], [o] and [u] between F1 and F2, as well for [e] between F2 and F3). Results also showed that there are statistically significant correlations between the fundamental frequency and formants, but only within female speakers of the Serbian language (between F0 and F1 and F0 and F3). Within both male and female speakers of Serbian, the results showed significant correlations between F1 and F2 for the front vowel [i] and back vowels [u] and [o]. In this dissertation, the significance of the correlation of F2 and F3 in pronunciation of the front vowel [i] was also confirmed. Numerous authors (Chistovich & Lublinskaya, 1979; Chistovich, 1985; Harrington, 2013, etc.) wrote about their role and the importance of their relations in the perception of vowel quality. Broadly speaking, these tests revealed that there are less significant correlations among formants within speakers of Croatian, than within speakers of Serbian. Further analysis showed a greater number of significant correlations for the front vowel [i]. If we compare the ratio of significant correlations of F1-F2 and F2-F3 relations, the results show more statistically significant correlations between F1 and F2. In the end, it should once again be emphasized that this study sought to contribute to the phonetics of the Croatian language by determining the reference formant frequencies of the Croatian vowel system. The sociophonetic contribution of the research is reflected in the contrastive analysis of the Croatian and Serbian languages, which are laid down in certain acoustic differences between the analysed languages, as well as in a study of sexual dimorphism of some acoustic parameters. Also, it is expected that the reference formant frame will be useful for evaluations of suspects’ pronuntiation in real forensic cases. In the future, it is expected that in the field of Croatian forensic phonetics, more complex acoustic analyses will be used, according to relevant and recent forensic research standards, which include different types of formant analysis. Moreover, it is assumed that this dissertation can contribute to the acoustic description of regional varieties of the Croatian language, which would be placed into the context of the Croatian RP.

Actions (login required)

View Item