Formant Distribution Research Articles

Do long-term acoustic-phonetic features and mel-frequency cepstral coefficients provide complementary speaker-specific information for forensic voice comparison?

A growing number of studies in forensic voice comparison have explored how elements of phonetic analysis and automatic speaker recognition systems may be integrated for optimal speaker discrimination performance. However, few studies have investigated the evidential value of long-term speech features using forensically-relevant speech data. This paper reports an empirical validation study that assesses the evidential strength of the following long-term features: fundamental frequency (F0), formant distributions, laryngeal voice quality, mel-frequency cepstral coefficients (MFCCs), and combinations thereof. Non-contemporaneous recordings with speech style mismatch from 75 male Australian English speakers were analyzed. Results show that 1) MFCCs outperform long-term acoustic phonetic features; 2) source and filter features do not provide considerably complementary speaker-specific information; and 3) the addition of long-term phonetic features to an MFCCs-based system does not lead to meaningful improvement in system performance. Implications for the complementarity of phonetic analysis and automatic speaker recognition systems are discussed.

Acoustic and Spectral Analysis of Adi Triphthongs

Adi, a zero-resource indigenous tribal language of Arunachal Pradesh, originated in Tibeto-Burman. According to the “UNESCO Atlas of the World's Languages in Danger of 2017” report, Adi is critically endangered. In this research, two new Adi triphthongs, /uai/ [uai] and /oai/ [ɔai], are proposed in the phoneme set of the Adi language. Adi has 51 phonemes, including 14 monophthongs (7 short and 7 long vowels), 19 diphthongs, two triphthongs, and 16 consonants. This research analysed Adi triphthongs’ spectral and acoustic properties, including intensity variation, long-term average spectrum (LTAS), and formant distribution. The mean of the first four formant frequencies of male and female Adi speakers is measured, which may be the identification of these two phones in the Adi speech recognition systems. Four different Adi words for each triphthong are recorded from 68 individual native Adi speakers (32 males and 36 females) in Arunachal Pradesh. This study may be helpful for phonetic modelling to build speech recognition applications for the Adi language.

Monozygotic twin sisters differ in voice quality: Acoustic analysis of the diphthong [‘oj] produced by Brazilian native speakers

Identical twins challenge individual voice recognition applied in communication, criminal investigation, and health technologies. The similarities between siblings are due to the genetic heritage that produced either their vocal tract or the psychological processes acting since birth. To contribute to a precise individual identification by voice, we tested the hypothesis that even MZ siblings would differ in the spectral distribution of formants and sound quality. We recorded 86 same-sex twin pairs (71 MZ, 15 DZ 30 ± 12 years old) saying the Portuguese verse of “Hi, my name is (Pedro or Ana).” We trimmed the diphthong Ój (“Hi”) and measured f0, F1-F4 formants, jitter, shimmer, and Harmonic-to-noise ratio (HNR). As the male sampling was small, we compared only female twins using the Wilcoxon-matched test, the intrapair Correlation Coefficient (ICC), and GLMM models. On average, [Ój] lasted 0.24 ± 0.07 s, and f0 ranged from 469 to 510 kHz according to sex. We found no differences between the MZ or DZ sisters in any acoustic parameter, besides F1 for MZ ones. Also, MZ sisters were more alike (F1, jitter, HNR, weak; F3, F4 moderate) than DZ (only F1 high ICC). The interaction zygosity × age affected intrapair similarity, the older the sisters, the greater their differences. Short voice segments, like the diphthong [Ój], may not be good for differentiating identical twins but provide insights to discussion; F2, F4, and shimmer should be investigated in future studies considering males, younger participants, and other vocal expressions.

Категория падежа в ваховском хантыйском языке (на материале современных полевых данных базы LingvoDoc)

Introduction. In Khanty dialects, the category of grammatical case shows its largely heterogeneous character and varies in terms of quantity and quality. As for Eastern Khanty, linguists hold different views on the composition of this category and its representatives, which results from a discrepancy in interpreting statuses of some case indicators. Goals. The article seeks to systematize available data on case indicators identified in Vakh Khanty and presented in scholarly studies of the nineteenth-twentieth centuries, compare them to field data obtained in the twenty first century and digitized on the LingvoDoc platform for independent research. Methods. The analysis of published data on Vakh Khanty has been carried out via the descriptive/comparative and structural research methods. Field data collection techniques for Vakh Khanty included direct observations of native speakers’ speech, a survey, interviews on various topics, audio recordings, and transcriptions of the latter. Results. Our review of the scholarly literature from the past centuries on the Vakh Khanty dialect shows a variety of opinions about the number of attested case markers and their functional specifics, an inconsistency of approaches to terming some of them, and a variance of the morphemic form designations. Analysis of the contemporary field data with the LingvoDoc platform’s functional tools and comparison of the results against some previously described data do confirm a stability of the Vakh Khanty dialect case system, identify the source of terminological synonymy in the nominal paradigm, and develop a classification of case markers based on their functional features. The functional parameter makes it possible to distinguish between a case marker with a predominantly grammatical orientation and a group of case markers whose profile is associated with the expression of semantic relations. In the group of semantic cases, further division into subgroups is also possible, due to the type of semantic specialization of individual case markers. Conclusions. Semantic cases fall into a subgroup of cases with spatial semantics, a subgroup with the semantics of compatibility, and a subgroup that combines cases with oblique (indirect) semantics. The distribution of case formants into subgroups is complicated by certain syncretism of semantics in a large number of case morphemes, which ultimately leads to some blurring of the subgroups’ boundaries and a variance in the terminological nomination of case morphemes.

V-to-V coarticulation may be extinct in the wild

Ever since the results of [S. E. G. Öhman, “Coarticulation…,” J. Ascoust. Soc. Am. 39, 151 (1966)] that a vowel could affect the preceding one, the phenomenon of anticipatory VCV coarticulation has been a subject of study. Most results establishing these effects have been obtained from curated laboratory speech [e.g., M. Grosvald, “Interspeaker variation…,” J. Phonetics 37, 173 (2009)], showing anticipatory V-to-V coarticulation across one, two and three syllables in English utterances. Here, we attempted to replicate these findings “in the wild” using the Buckeye Corpus of conversational English. F1 and F2 of [ʌ] or [ə] tokens were obtained in all [ʌ/ə]C(C…)V contexts. Two-dimensional Kolmogorov-Smirnoff tests were carried out to compare the F1/F2 distributions of these tokens in different contexts. The formant distributions of [ʌ/ə] in specific vowel contexts were also compared with the total formant distributions of [ʌ/ə] in all contexts. The replication effort was a failure. Essentially no significant differences of the [ʌ/ə] formant distributions could be found among the data, no matter how extreme the difference of the following vowel (e.g., [i] vs [ɑ] contexts). This poses a problem for theories of speech perception which rely on such effects.

Stability, reliability, and sensitivity of acoustic measures of vowel space: A comparison of vowel space area, formant centralization ratio, and vowel articulation index.

Vowel space (VS) measurements can provide objective information on formant distribution and act as a proxy for vowel production. There are a number of proposed ways to quantify vowel production clinically, including vowel space area, formant centralization ratio, and vowel articulation index (VAI). The stability, reliability, and sensitivity of three VS measurements were investigated in two experiments. Stability was explored across three inter-recording intervals and challenged in two sensitivity conditions. Data suggest that VAI is the most stable measure across 30 s, 2 h, and 4 h inter-recording intervals. VAI appears the most sensitive metric of the three measures in conditions of fatigue and noise. These analyses highlight the need for stability and sensitivity analysis when developing and validating acoustic metrics, and underscore the potential of the VAI for vowel analysis.

Effects of consonantal context on the learnability of vowel categories from infant-directed speech.

Recent studies have shown that vowels in infant-directed speech (IDS) are characterized by highly variable formant distributions. The current study investigates whether vowel variability is partially due to consonantal context, and explores whether consonantal context could support the learning of vowel categories from IDS. A computational model is presented which selects contexts based on frequency in the input and generalizes across contextual categories. Improved categorization performance was found on a vowel contrast in American-English IDS. The findings support a view in which the infant's learning mechanism is anchored in context, in order to cope with acoustic variability in the input.

Acoustic Analysis of Intensive Language Exposure on Second Language Learners’ Cognitive Perception Pattern of American English Phonemes

This paper attempts to quantify the perception of tense vowel /i/ and lax vowel /I/ in American English by native speakers and Chinese learners and testify if Chinese learners also rely more on acoustic features than duration to differentiate the minimal pair /i/-/I/ in the phonetic system of American English. Based on the related phonetic studies in second language learners (SLLs), the perceptual assimilation model (PAM) and the speech learning model (SLM) were adopted to discuss the effect of intensive language exposures on the perception of English phonemes among Chinese learners. To quantify the effects of the ILE on the perception pattern of the SLLs, two English phonetic continuums were constructed for a tense-lax vowel contrast /i/ and /I/. Then, three groups of subjects, involving 60 English major students and 10 native speakers, received two identification tests and one discrimination test. The test results reveal that the native speakers and the SLLs relied on different acoustic cues to distinguish between /i/ and /I/. The former mainly depended on formant distribution, while the latter on duration. Besides, it is also concluded that the ILE enabled the SLLs to develop a similar perception pattern to that of the native speakers. The research sheds new light on the acquisition of second language among Chinese learners.

Influence on tone perception from vowels with different formant distributions

Influence on tone perception from vowels with different formant distributions

若年層韓国語ソウル方言における/o/と/u/の言語変化の一側面―先行子音種と発話単位の違いが2母音の重なりに与える影響―

若年層韓国語ソウル方言における/o/と/u/の言語変化の一側面―先行子音種と発話単位の違いが2母音の重なりに与える影響―

Text-independent speech emotion recognition using frequency adaptive features

In this paper a text-independent emotional speech feature extraction method is studied based on various spectral frequency bands of speech formants. First, the speech emotional feature analysis is performed for different text contents. Various sentences are involved to study the phonetic influences. Formant frequencies are grouped into different classes in order to reduce the text variability. These speech features are sensitive to phonetic changes in the sentence. Speaker emotions are then modeled in different formant groups. Second, adaptive fundamental frequency and Teager Energy Operator are constructed in different frequency bands. The Teager frequency bands are dynamically adapted to different pitch and formant distributions. The proposed adaptation is sensitive to emotional changes in speech, as shown in statistics of mean, variance, maximum and minimum. Statistics on the basic acoustic parameters are used as the emotional features. Experimental results show that the proposed emotional features are robust against text changes with the lowest variance value of 0.034. The final recognition results for six major emotion types are improved constantly and 5 percent improvement for sadness and 3.3 percent improvement for boredom are observed.

Perceived vocal attractiveness by gay listeners in Hong Kong

This study investigates how Cantonese-speaking homosexual males perceive vocal attractiveness; namely, how different phonetic properties of voice influence their judgment. To this end, the present project follows the methodology in Xu et al’s (2013) study of vocal attractiveness and emotion in English, and recruited 14 Cantonese-speaking gay men in Hong Kong to rate the attractiveness of digitally synthesised stimuli. The stimuli varied in voice quality (breathy / modal / pressed), fundamental frequency and formant dispersion, all of which were means to project the speaker’s intended body size. The attractiveness ratings show that homosexual male subjects prefer male voices with low pitch, breathy voice quality and normal (neither dense nor sparse) distribution of formants. Interestingly, these preferences resemble the ratings by female English subjects in Xu et al’s (2013) study with one exception—the females favoured male voices with densely distributed formants, which signal a large body. This divergen...

Universal vs. language-specific aspects in human vocal attractiveness: An investigation towards Japanese native listeners’ perceptual pattern

Studies on Western societies show that male voices with acoustic parameters encoding a big body size (low F0, narrow formant dispersion, and F0 range) were considered to be attractive, while the opposite was true for female voices (e.g., [Xu et al., 2013, PLoS ONE, 8(4), e62397]). The present work investigates whether Japanese native listeners are guided by the same principles in assessing the voices of the opposite sex. We replicated the design in Xu et al. (2013) with the added parameter of creaky voice, which is prevalent in North America nowadays and hotly debated in terms of attractiveness. Thirty-four heterosexual participants (16 female) rated the attractiveness of synthetic stimuli controlling for F0 height, formant distribution, F0 range and voice quality. Results indicate that their preferences for voice quality are similar with studies on Western societies (breathy > modal >creaky >pressed/tensed). Additionally, low-pitched male voice with narrow formant dispersion and high-pitched female voice...

A Chinese alligator in heliox: formant frequencies in a crocodilian.

ABSTRACTCrocodilians are among the most vocal non-avian reptiles. Adults of both sexes produce loud vocalizations known as ‘bellows’ year round, with the highest rate during the mating season. Although the specific function of these vocalizations remains unclear, they may advertise the caller's body size, because relative size differences strongly affect courtship and territorial behaviour in crocodilians. In mammals and birds, a common mechanism for producing honest acoustic signals of body size is via formant frequencies (vocal tract resonances). To our knowledge, formants have to date never been documented in any non-avian reptile, and formants do not seem to play a role in the vocalizations of anurans. We tested for formants in crocodilian vocalizations by using playbacks to induce a female Chinese alligator (Alligator sinensis) to bellow in an airtight chamber. During vocalizations, the animal inhaled either normal air or a helium/oxygen mixture (heliox) in which the velocity of sound is increased. Although heliox allows normal respiration, it alters the formant distribution of the sound spectrum. An acoustic analysis of the calls showed that the source signal components remained constant under both conditions, but an upward shift of high-energy frequency bands was observed in heliox. We conclude that these frequency bands represent formants. We suggest that crocodilian vocalizations could thus provide an acoustic indication of body size via formants. Because birds and crocodilians share a common ancestor with all dinosaurs, a better understanding of their vocal production systems may also provide insight into the communication of extinct Archosaurians.

Impact of mobile phone usage on speech spectral features: some preliminary findings

The manner of using a mobile phone in voice communications can significantly affect the spectral characteristics of the speech signal. This article presents preliminary results of the analysis of the long-term average speech spectrum (LTASS), the long-term formant distribution (LTF) in voiced sounds, and vowel formants F1, F2 and F3 of six speakers in five modes of mobile phone usage. These modes are: normal holding of a mobile phone (NOR), with a bonbon (sweet) in the mouth (BON), with a cigarette between the lips (CIG), with the mobile phone between cheek and shoulder (SHO) and with the hand covering the mobile phone and mouth (HAN). The results show that each mode has an impact on spectral features and that the modes HAN and SHO have the greatest impact. The most striking results are the relative displacement of F1, which can reach 30% (e.g. vowel /a/ in HAN mode for males), formant F2, near 15% (vowel /i/ in SHO mode for males), and formant F3, about 5% (vowel /u/ in CIG mode for females). These findings suggest that forensic practitioners should exercise caution in interpreting formant measurements in speaker identification cases involving mobile phone transmission.

한국어 자연발화 음성코퍼스의 남녀 모음 포먼트 비교 연구

The aim of this work is to compare the vowel formants of the ten adult female speakers in their twenties and thirties from the Seoul corpus[7] with those of corresponding Korean male speakers from the same corpus and of American female speakers from the Buckeye corpus[4]. In addition, various linguistic factors that are expected affect the formant frequencies were examined to account for the distribution of the vowel formants. Formant frequencies extracted from the Seoul corpus were also compared to those from read speech. The results showed that the formant distribution of the spontaneous speech was very different from that of the read speech, while the comparison between the female and male speakers was similar in both languages. To a greater or lesser degree, the potential linguistic factors influenced the formant frequencies of the vowels.

Choice of Model of Test Signal at an Assessment of Security of Speech Information from Leakage through Technical Channels

Revaluation of the basic dependences which are used in the method of an assessment of security of speech information from leakage through technical channels is made. The statement and results of experiment for definition of formant distribution for cases of the usual and forced speech are described. Using these distributions it is possible to define contributions of frequency bands to total legibility of speech. Dependence of word legibility on formant legibility for a case of forced speech is received. The paper presents an experiment to determine amplitude structure of the speech by results of which conclusions of sufficient levels of a test signal are drawn for an assessment of security of speech information.

Examining correlations between phonetic parameters: Implications for forensic speaker comparison

The research presented in this paper builds upon a previous pilot study (Gold and Hughes 2012). This paper explores the correlation structure of speech parameters from a sociolinguistically homogeneous set of male speakers of Southern Standard British English using a series of segmental, suprasegmental and linguistic parameters. Data was extracted from a subset of speakers from the Dynamic Variability in Speech (DyViS) database (Nolan et al., 2009) and consist of: midpoint F1, F2 & F3 values for /a ɔ u/, midpoint F1, F2 & F3 values hesitation markers UM and UH, dynamic F1, F2 & F3 values for PRICE /aɪ/, long-term formant distributions (LTFD) F1-F4, mean and standard deviation of fundamental frequency (F0), mean articulation rate (AR), voice onset time (VOT) for word-initial /t/ and /k/, and click rate (ingressive velaric stops). The results of the study present a complex correlation structure between linguistic-phonetic variables, and not all correlations are predicted by phonetic theory. The results of the correlations are discussed in relation to implications that exist when combining parameters for forensic speaker comparison casework; specifically, the caution that needs to be yielded by experts in casework to avoid over- or under-estimating the strength of evidence.

Feedback-driven corrective movements in speech in the absence of altered feedback

Altered auditory feedback often evokes a compensatory vocal response in speakers, providing evidence that errors in speech production can be rapidly corrected online. To assess whether the same compensatory mechanism is employed in natural, unaltered speech, we carried out an acoustic analysis characterizing formant movement during single speaking trials. Subjects produced 200 repetitions each of three different monosyllabic words in the MEG scanner and, separately, in the presence of varying background noise levels. To assess corrective responses, we compared the centricity of formant values at the beginning to that at the middle of each trial. In all subjects, we found strong evidence of vowel “centering”—that is, a corrective movement mid-utterance that caused utterances at the periphery to move closer to the center of the formant distribution. Across subjects, the magnitude of vowel centering was correlated with auditory cortical suppression preceding the corrective movement, suggesting that the suppression may serve as a neural mechanism for error detection and correction. These findings suggest that less-prototypical utterances, which make up a large proportion of natural speech, are processed as potential errors, and that feedback-driven speech error correction is occurring constantly on a small scale.

Gender Categorization in Cochlear Implant Users

In this study, the authors examined the ability of subjects with cochlear implants (CIs) to discriminate voice gender and how this ability evolved as a function of CI experience. The authors presented a continuum of voice samples created by voice morphing, with 9 intermediate acoustic parameter steps between a typical male and a typical female. This method allowed for the evaluation of gender categorization not only when acoustical features were specific to gender but also for more ambiguous cases, when fundamental frequency or formant distribution were located between typical values. Results showed a global, though variable, deficit for voice gender categorization in CI recipients compared with subjects with normal hearing. This deficit was stronger for ambiguous stimuli in the voice continuum: Average performance scores for CI users were 58% lower than average scores for subjects with normal hearing in cases of ambiguous stimuli and 19% lower for typical male and female voices. The authors found no significant improvement in voice gender categorization with CI experience. These results emphasize the dissociation between recovery of speech recognition and voice feature perception after cochlear implantation. This large and durable deficit may be related to spectral and temporal degradation induced by CI sound coding, or it may be related to central voice processing deficits.