Estimation Of Vocal Tract Length Research Articles

An acoustic study on age-related changes in vowel production of Chinese

This paper investigates the relationship between vowel production and age using speech data from 109 Chinese L1 speakers (61 females and 48 males) covering an age range of 20 to 80 years. Acoustical estimation of vocal tract length (VTL) as well as multiple acoustic metrics are analyzed with generalized additive mixed models (GAMM). The results indicate that: (1) After controlling for VTL, F0 and duration, vowels show a centralization trend with increasing age, with a more significant effect observed in female speakers; (2) VTL does not significantly change with age; (3) The patterns observed in vowel distinctiveness and duration may present evidence contradicting the notion of vowel lengthening as a compensatory mechanism; (4) The patterns of age-related changes in different measurements and different genders are diverse. The U-shaped change patterns are found in the male speakers and the age around 50 may serve as a turning point. Based on these findings, we have explored some possible reasons for inconsistent conclusions in previous studies. The physiological aging phenomena of vowel production and potential compensatory mechanisms on motor control abilities, as well as other possible influencing factors, are also discussed.

Inaccurate but predictable: Vocal-tract length estimation and gender stereotypes in height perception

Research suggests human listeners are not very accurate in assessing the size of adults from their speech, though they appear to be consistent in their judgments across listeners. Two experiments were conducted to investigate the importance of the higher formants for providing consistent height judgments, how consistent these height judgments are across replications, and the role of f0 and social knowledge in maintaining the stability of apparent speaker height judgments. Listeners were presented with syllables produced by 30 adult male and female speakers, and were asked to identify the word, the gender of the speaker, and the height of the speaker. In a second experiment, listeners were presented with voiced and (synthetic) whispered speech and asked to provide the same responses. Results indicate that speakers use acoustic cues in largely predictable ways, leading to consistent apparent height judgments when averaged across listeners. However, the behavior of individual listeners is unpredictable, and the accuracy of apparent height judgments with respect to veridical heights is low. Finally, results suggest that non-acoustic social knowledge regarding the expected sizes of adult males and females play an important role in the determination of apparent height from speech.

MFCC Parameters of the Speech Signal: An Alternative to Formant-Based Instantaneous Vocal Tract Length Estimation

On the one hand, the relationship between formant frequencies and vocal tract length (VTL) has been intensively studied over the years. On the other hand, the connection involving mel-frequency cepstral coefficients (MFCCs), which concisely codify the overall shape of a speaker’s spectral envelope with just a few cepstral coefficients, and VTL has only been modestly analyzed, being worth of further investigation. Thus, based on different statistical models, this article explores the advantages and disadvantages of the latter approach, which is relatively novel, in contrast to the former which arises from more traditional studies. Additionally, VTL is assumed to be a static and inherent characteristic of speakers, that is, a single length parameter is frequently estimated per speaker. By contrast, in this paper we consider VTL estimation from a dynamic perspective using modern real-time Magnetic Resonance Imaging (rtMRI) to measure VTL in parallel with audio signals. To support the experiments, data obtained from USC-TIMIT magnetic resonance videos were used, allowing for the 2D real-time analysis of articulators in motion. As a result, we observed that the performance of MFCCs in case of speaker-dependent modeling is higher, however, in case of cross-speaker modeling, which uses different speakers’ data for training and evaluating, its performance is not significantly different of that obtained with formants. In complement, we note that the estimation based on MFCCs is robust, with an acceptable computational time complexity, coherent with the traditional approach.

Vocal Tract Length Estimation Using Accumulated Means of Formants and Its Effects on Speaker-Normalization

Differences in vocal tract lengths (VTLs) in individual speakers cause variations in acoustic features of phonemes. In this paper, a simple method to estimate speaker-specific VTLs and to quantitatively evaluate some speaker-normalization effects of the VTLs is proposed. We employed accumulated means of formant trajectories to estimate the VTLs of speakers ranging from children to adults. For the formant estimation, the inverse-filter control (IFC) system was used. In the system, the decision of analysis order, which means number of formants to be estimated, is automated. Moreover, to evaluate the speaker-normalization effect of VTLs, we proposed the data reduction method, which can reasonably find dense areas of ellipses from distributions in the formant space. Using these ellipse areas, we evaluated the three normalization effects of VTLs: normalization by the mean of all VTLs as the standard, by speaker-categorical means of VTLs, and by individual VTLs. The area reduced from the standard area of the original data by 39.5% and 46.6% in the case of the categorical means and individual VTLs, respectively. As a result, our proposed method was used to provide a “normalized vowel map (NVM)” that visualizes universal vowel-distributions as a core image of linguistic information. Finally, we compared the estimated VTLs with those by another method based on magnetic resonance imaging (MRI) data, using the proposed methods.

Towards higher precision in vocal tract length estimation

Approaches to vocal tract length (VTL) estimation typically differ in how much credibility is allocated to certain formants as predictors of length, but are alike in allocating equal credibility to all vowel spectra. The latter may be problematic, as asymmetries in the phonotactic frequency of certain vowels and measurement errors for higher formants can influence VTL estimates. Herein, an additional parameter is proposed which privileges vowel spectra that approximate a uniform tube. The proposed metric for this proximity is standard variance (σ) in Φ, where Φ=Fn/(2n-1), n represents the integer label of the formant, and σΦ for a uniform tube is 0. The five estimators detailed in Lammert & Narayanan (2015) were tested on the speech of two adult males. Each of the estimators were run using all vowel spectra as well as only those for which σΦ < 50 Hz. Estimates from all vowel spectra had σVTL well above 1 cm, which is undesirably large, given that the adult vocal tract only ranges from 13 to 20 cm. However, estimates from vowel spectra with σΦ < 50Hz had σVTL less than 1 cm. This higher precision is especially striking since spectra with σΦ < 50 Hz constituted a small fraction of the total.

An investigation of the systematic use of spectral information in the determination of apparent-talker height

The perception of apparent-talker height is mostly determined by the fundamental frequency (f0) and spectral characteristics of a voice. Although it is traditionally thought that spectral cues affect apparent-talker height by influencing apparent vocal-tract length, a recent experiment [Barreda (2016). J. Phon. 55, 1-18] suggests that apparent-talker height can vary significantly within-talker on the basis of phonemically-determined spectral variability. In this experiment, listeners were asked to estimate the height of 10 female talkers based on manipulated natural productions of bVd words containing one of /i æ ɑ u ɝ/. Results indicate that although listeners appear to use vocal-tract length estimates in determining apparent-height, apparent-talker height also varies significantly within-talker based on the inherent spectral and source characteristics of different vowels, with vowels with lower formant-frequencies and f0 being associated with taller talkers overall. The use of spectral and f0 information in apparent-height estimation varied considerably between listeners, resulting in additional variation in the apparent-height of talkers. Although the use of acoustic information in the determination of apparent-height was highly systematic, it does not necessarily follow from the empirical relationship between speech acoustics and actual talker height.

Using CT to predict vocal tract resonances in mammal vocalizations: Are calls nasalized?

Males of several species of deer have a descended larynx, which gives them an unusually long vocal tract (VT). They can extend their VT by further lowering their larynx during the production of their sexual loudcalls. Formant frequencies are lowered as the vocal tract is extended, as predicted when approximating the vocal tract as a uniform quarter wavelength resonator. However, formant frequencies in polygynous deer follow uneven distribution patterns, suggesting that the vocal tract shape is in fact rather complex. We CT-scanned the artificially extended vocal tract of two adult fallow deer, and measured the cross-sectional area of the supralyngeal vocal tract along the oral and nasal pathways. We used this data to model resonances patterns produced by these VT including the oral pathway, the nasal pathway, or both. We found that the combined oral/nasal VT produced a formant pattern that more closely matches that observed in fallow deer groans and enables a better estimation of VT length from formant pa...

Voice parameters predict sex-specific body morphology in men and women

Studies of several mammalian species confirm that formant frequencies (vocal tract resonances) predict height and weight better than does fundamental frequency (F0, perceived as pitch) in same-sex adults due to differential anatomical constraints. However, our recent meta-analysis (Pisanski et al., 2014, Animal Behaviour, 95, 89–99) indicated that formants and F0 could explain no more than 10% and 2% of the variance in human height, respectively, controlling for sex and age. Here, we examined whether other voice parameters, many of which are affected by sex hormones, can indicate additional variance in human body size or shape, and whether these relationships differ between the sexes. Using a cross-cultural sample of 700 men and women, we examined relationships among 19 voice parameters (minimum–maximum F0, mean F0, F0 variability, formant-based vocal tract length estimates, shimmer, jitter, harmonics-to-noise ratio) and eight indices of body size or shape (height, weight, body mass index, hip, waist and chest circumferences, waist-to-hip ratio, chest-to-hip ratio). Our results confirm that formant measures explain the most variance in heights and weights of men and women, whereas shimmer, jitter and harmonics-to-noise ratio do not indicate height, weight or body mass index in either sex. In contrast, these perturbation and noise parameters, in addition to F0 range and variability, explained more variance in body shape than did formants or mean F0, particularly among men. Shimmer or jitter explained the most variance in men's hip circumferences (12%) and chest-to-hip ratios (6%), whereas harmonics-to-noise ratio and formants explained the most variance in women's waist-to-hip ratios (11%), and significantly more than in men's waist-to-hip ratios. Our study represents the most comprehensive analysis of vocal indicators of human body size to date and offers a foundation for future research examining the hormonal mechanisms of voice production in humans and perceptual playback experiments.

On Short-Time Estimation of Vocal Tract Length from Formant Frequencies.

Vocal tract length is highly variable across speakers and determines many aspects of the acoustic speech signal, making it an essential parameter to consider for explaining behavioral variability. A method for accurate estimation of vocal tract length from formant frequencies would afford normalization of interspeaker variability and facilitate acoustic comparisons across speakers. A framework for considering estimation methods is developed from the basic principles of vocal tract acoustics, and an estimation method is proposed that follows naturally from this framework. The proposed method is evaluated using acoustic characteristics of simulated vocal tracts ranging from 14 to 19 cm in length, as well as real-time magnetic resonance imaging data with synchronous audio from five speakers whose vocal tracts range from 14.5 to 18.0 cm in length. Evaluations show improvements in accuracy over previously proposed methods, with 0.631 and 1.277 cm root mean square error on simulated and human speech data, respectively. Empirical results show that the effectiveness of the proposed method is based on emphasizing higher formant frequencies, which seem less affected by speech articulation. Theoretical predictions of formant sensitivity reinforce this empirical finding. Moreover, theoretical insights are explained regarding the reason for differences in formant sensitivity.

Vocal indicators of body size in men and women: a meta-analysis

Animals often use acoustical cues, such as formant frequencies, to assess the size of potential mates and rivals. Reliable vocal cues to size may be under sexual selection. In most mammals and many other vertebrates, formants scale with vocal tract length allometrically and predict variation in size more reliably than fundamental frequency or pitch (F0). In humans, however, it is unclear from previous work how well voice parameters predict body size independently of age and sex. We conducted a meta-analysis to establish the strength of various voice–size relationships in adult men and women. We computed mean weighted correlations from 295 coefficients derived from 39 independent samples across five continents, including several novel and large cross-cultural samples from previously unpublished data. Where possible, we controlled for sample size, sample sex, mean age, geographical location, study year, speech type and measurement method, and ruled out publication bias. Eleven of 12 formant-based vocal tract length (VTL) estimates predicted men's and women's heights and weights significantly better than did F0. Individual VTL estimates explained up to 10% of the variance in height and weight, whereas F0 explained less than 2% and correlated only weakly with size within sexes. Statistically reliable size estimates from F0 required large samples of at least 618 men and 2140 women, whereas formant-based size estimates required samples of at least 99 men and 164 women. The strength of voice–size relationships varied by sample size, and in some cases sex, but was largely unaffected by other demographic and methodological variables. We confirm here that, analogous to many other vertebrates, formants provide the most reliable vocal cue to size in humans. This finding has important implications for honest signalling theory and the capacity for human listeners to estimate size from the voice.

On instantaneous vocal tract length estimation from formant frequencies

The length of the vocal tract and its relationship with formant frequencies is examined at fine temporal scales with the goal of providing accurate estimates of vocal tract length from acoustics on a spectrum-by-spectrum basis despite unknown articulatory information. Accurate vocal tract length estimation is motivated by applications to speaker normalization and biometrics. Analyses presented are both theoretical and empirical. Various theoretical models are used to predict the behavior of vocal tract resonances in the presence of different vocal tract lengths and constrictions. Real-time MRI with synchronized audio is also utilized for detailed measurements of vocal tract length and formant frequencies during running speech, facilitating the examination of short-time changes in vocal tract length and corresponding changes in formant frequencies, both within and across speakers. Previously proposed methods for estimating vocal tract length are placed within a coherent framework, and their effectiveness is evaluated and compared. A data-driven method for VTL estimation emerges as a natural extension of this framework, which is then developed and shown to empirically outperform previous methods on both synthetic and real speech data. A theoretical justification for the effectiveness of this new method is also explained. [Work supported by NIH.]

Continuous estimation of VTL from vowels using a linearly VTL-covariant speech feature

Vowel sounds encode information about the speaker's vocal tract length (VTL) as well as vowel type. This paper shows how VTL can be estimated on a frame-by-frame basis from vowels to facilitate speaker tracking and make automatic speech recognition (ASR) more robust in multi-speaker environments. The algorithm is based on a new, VTL-covariant speech feature that encodes the size information linearly. The paper shows that this new speech feature is better suited for VTL estimation than the more traditional cepstral coefficients. The VTL estimation is based on a Gaussian mixture model that was trained on the speech material of speakers with known body height but unknown VTL. Research supported by the Austrian Science Fund (FWF) [J2541-N15], EOARD [FA8655-05-1-3043], and the UK-MRC [G0500221].

VTLN Through Frequency Warping Based on Pitch

This article describes a Vocal Tract Length Nor­ malization (VTLN) procedure through frequency warping based on pitch estimates. This procedure aims to reduce the inter-speaker variability of speech signals in order to obtain a robust automatic speech recognition system. Two additional methods are also described: one for reducing the environment variability and another for compensating the coarticulation effects on connected word pronunciation. En­ vironment variability is compensated by explicitly modeling some frequent noise phenomena. Coarticulation phenomena compensation reduces speech signal variability by modeling events that result from coarticulation between adjacent mod­ els. Inter-speaker variability removal is performed by a traditional speaker normalization method, which consists in expanding or compressing the Mel filterbank bandwidths, in order to normalize the Vocal Tract Length (VTL) of each speaker. Most of the existing methods for VTL estimation are based on formant estimation, but the difficulty of formant estimation is a known performance limitation. The proposed method over­ comes such a problem since it estimates the warping factor through pitch. The recognition results, obtained for a tele­ phone digit recognition task (with phones and sub words as units), prove that this procedure leads to similar improve­ ments to those obtained with traditional methods based on formant estimates, actually outperforming them in some sit­ uations.

A comparison between vocal tract length estimates from acoustic and x-ray data

Thirty vocal tract length measurements from lip corners to glottis were obtained from cine-x-ray sagittal views and lateral-frontal photographic data. Material for two French speakers includes vowels [i, ɛ, ɑ, ɔ, u] in three symmetric [ə CVC] frames where C is either [b, d] or [g]. Four formant frequencies were measured from the simultaneous audio recording using covariance LPC on closed glottis sections, and were corrected for the frequency effects of lip radiation and wall vibrations. Calculated bandwidths from Fant's data and the modified formant frequencies were used to derive area functions for several tract lengths. Use of Paige and Zue's criterion [IEEE-AU 18, 268–270 (1970)] of minimum perturbation with respect to a uniform tube yielded close agreement between measured and estimated lengths for all vowels except [u]. Removal of the last (lip) section in criterion calculations gives better results for [u].

Estimation of vocal-tract shapes from acoustical analysis of the speech wave: The state of the art

The state of the art of the method for estimating vocal-tract area functions based on the linear prediction of speech is discussed. Limitations and problems involved in this method are: 1) frequency band limitation of the signal, 2) source characteristics, 3) boundary conditions, 4) energy losses within the vocal tract, 5) vocal-tract length estimation, 6) dynamics of the vocal tract, and 7) evaluation of the method [1]. The proposed solutions for these problems are discussed together with some of the current developments pertaining to these problems. Advantages and disadvantages of this method are also discussed in comparison with the lip impulse response method.