Vocal Tract Area Research Articles

An overview of the physiology, physics and modeling of the sound source for vowels.

The vibration of the vocal folds produces the primary sound source for vowels. This paper first reviews vocal fold anatomy and the kinematics associated with typical vibratory motion. A brief historical background is then presented on the basic physics of vocal fold vibration and various efforts directed at mathematical modeling of the vocal folds. Finally, a low-dimensional model is used to simulate the vocal fold vibration under various conditions of vocal tract loading. In particular, a “no-tract” case is compared to two cases in which the voice source is coupled to vocal tract area functions representing the vowels /i/ and /a/, respectively.

Perception of Lateral Misarticulation and Its Physical Correlates

In order to clarify the relationship between perceptual diagnosis of lateral misarticulation (LM) by sophisticated listeners and the physical correlates of LM, three experiments using sustained speech /∫/ were conducted. Experiment 1 was designed to compare the spectral envelopes of normal speech (NS) /∫/ with those of LM /∫/. Experiment 2 was designed to collect the auditory impressions of sophisticated listeners listening to LM and NS /∫/ with specific spectral envelope bands replaced by LM. These two experiments showed that: (1) the spectral envelopes of LM are flat or decrease along the frequency axis in the frequency band above approximately 4 kHz, and there is a substantial peak at around 3.2 kHz in LM, which varies peculiarly with time; (2) the replacement of the spectral envelope between 2.5 and 4.5 kHz of NS with that of LM resulted in a remarkable increase in auditory impressions of LM. The facts suggest that the spectral envelope characteristic of LM has a peculiar variation at around 3.2 kHz. Additionally, experiment 3 estimated the spectrum of sustained speech /∫/ using vocal tract area functions. The results suggest that typical peaks of LM are related to the length and position of the vocal tract constriction region.

A distinctive region model based on empirical vocal tract area functions

The development of the Distinctive Region Model (DRM) [Mrayati et al., Speech Commun. (1988)] is based on theoretically derived acoustic characteristics for a tube of uniform cross-sectional area assumed to approximate a neutral vocal tract configuration. Formant sensitivity functions calculated for the uniform tube are used to divide the vocal tract into distinctive regions that, when constricted or expanded, will cause the formant frequencies to change in a predictable pattern. This study compares the original DRM (based on a uniform tube) with a new version created from a neutral vocal tract area function derived from published MRI data. Because it is subject to physiologic constraints, this neutral area function is nonuniform in cross-sectional area variation but exhibits formant frequencies similar to a uniform tube. Sensitivity function calculations for F1, F2, and F3 also show similarities to those of a uniform tube, but the zero-crossing points that divide the vocal tract into distinctive regions are shifted toward the lips. The result is distinctive regions that are not symmetric about the vocal tract mid-point but rather the back and front regions occupy about 60% and 40% of the total tract length, respectively. [Work supported by NIH R01-DC04789.]

Perception of Vowel-to-Vowel Transitions with Different Formant Trajectories

In this paper, the perceptual effects of vowel-to-vowel transitions determined by different temporal variations of model parameters which specify the shapes of the vocal tract area function are investigated. It is shown that, (a) the method of deformation of the vocal tract area function between two targets can be perceptually important and (b) conversely, within certain limits, the time course of parameters from one state to another, and the precise synchronization of two parameters is not important for the correct identification of a vowel series. These characteristics are necessary but not sufficient to prove the existence of a phonetic gesture percept.

Speech synthesis by mapping articulator movement patterns to a shape-based area function model of the vocal tract

Orthogonal components of tongue and lip displacement data from two speakers in the Univ. of Wisconsin x-ray microbeam database (Westbury, UW–Madison, 1994) were determined with a principal components analysis (PCA). The PCA was performed on steady-state vowels and at the sentence level. For every time frame, the analysis consisted of fitting a cubic spline to four pellet points on the tongue and to modified positions for the incisor pellet and the upper and lower lip pellets. The modifications brought the pellet positions to the level of the airway. The PCA was performed over a collection of frames and showed that the first two (most significant in terms of variance) orthogonal components were similar in shape to those determined from a set of MRI-based vocal tract area functions [Story and Titze, J. Phonetics 26, 223–260 (1998)]. Next, the coefficients determined from a sentence-level PCA of the microbeam data were normalized and used as input to an area function model based also on orthogonal shaping components. The result is that articulatory movement patterns from data in the x-ray microbeam database can be transformed to movement patterns of a vocal tract area function to reproduce (synthesize) the original sentence. [Work supported by NIH R01-DC04789-01.]

Pithc-synchronous articulatory synthesis incorporated with the inverse solution of speech production

This paper presents a new proposal to synthesize natural sounds with less control parameters by combining the inverse speech production and pitch-synchronous articulatory synthesis. The pitch-synchronous excited Reflection-Type Line Analog (RTLA) model is employed as the synthesis filter. Multi-rate system sampling and dynamic scattering wave adjustment are used to handle the variable VT length and the acoustic continuity. The synthesizer is controlled by vocal-tract (VT) area functions. Given the targets of formant trajectories, the dynamic VT area function which is modeled by time variant VT length is derived using an inverse solution of speech production. A distinguishing feature of this method is that artificially specified formant trace can be precisely aimed in the synthetical sounds. Experimental results show that the formant target can be well matched by the synthetic sounds. Potential application to text-to-speech conversion of this method is discussed.

An auditory-feedback-based neural network model of speech production that is robust to developmental changes in the size and shape of the articulatory system.

The purpose of this article is to demonstrate that self-produced auditory feedback is sufficient to train a mapping between auditory target space and articulator space under conditions in which the structures of speech production are undergoing considerable developmental restructuring. One challenge for competing theories that propose invariant constriction targets is that it is unclear what teaching signal could specify constriction location and degree so that a mapping between constriction target space and articulator space can be learned. It is predicted that a model trained by auditory feedback will accomplish speech goals, in auditory target space, by continuously learning to use different articulator configurations to adapt to the changing acoustic properties of the vocal tract during development. The Maeda articulatory synthesis part of the DIVA neural network model (Guenther et al., 1998) was modified to reflect the development of the vocal tract by using measurements taken from MR images of children. After training, the model was able to maintain the 11 English vowel targets in auditory planning space, utilizing varying articulator configurations, despite morphological changes that occur during development. The vocal-tract constriction pattern (derived from the vocal-tract area function) as well as the formant values varied during the course of development in correspondence with morphological changes in the structures involved with speech production. Despite changes in the acoustical properties of the vocal tract that occur during the course of development, the model was able to demonstrate motor-equivalent speech production under lip-restriction conditions. The model accomplished this in a self-organizing manner even though there was no prior experience with lip restriction during training.

Noise source models for fricative consonants

Hybrid source models for fricative consonants are derived based on aeroacoustic principles of sound generation employing vocal tract area functions obtained from magnetic resonance imaging data of voiced and unvoiced English fricatives. Results based on data from a male and a female subject indicate that a linear source-filter model is fairly adequate for capturing essential spectral characteristics of sustained voiced and unvoiced strident fricatives below 10 kHz. The hybrid source models employ a combination of acoustic monopole and distributed dipole sources, and a voice source in the case of the voiced fricatives. The number of sources, source locations, and spectral characteristics and the relative source levels are chosen based on an analysis-by-synthesis approach and are motivated by aeroacoustic theory of speech production. The resulting model is computationally efficient and can be readily used for synthesis.

A new noniterative algorithm for computing acoustically constrained vocal tract area functions

We describe a new noniterative algorithm for generating the unique area function determined by the vocal tract length, the lip radius, and the spectral pair consisting of poles of the transfer function and zeros of the input impedance function. Our analysis is restricted to the class of piecewise-constant area functions defined on an even number of equal length intervals. The resulting algorithm involves fewer floating point operations per evaluation than the analogous method of Paige and Zue (1970).

A preliminary study of speech transformation using empirically defined articulatory modes

In previous work it has been shown that a speaker-specific set of vocal tract shapes (acquired using MRI) corresponding to vowels can be decomposed into orthogonal components or ‘‘modes,’’ and effectively parameterized by the modal coefficients. Furthermore, a nearly one-to-one mapping was found to exist between the modal coefficients of the two most significant modes and the F1−F2 formant space. Thus an inverse mapping of time-varying formants extracted from recorded speech back to modal coefficients (and consequently to vocal tract area functions) was made possible; the derived sequence of area functions can be used in a simulation of the original speech utterance. In this study, it is first shown that similar orthogonal modes exist for three additional speakers. It will then be demonstrated that the time-varying modal coefficients determined for a recorded sentence via the inverse mapping of one speaker can also be used to create a sequence of area functions based on one of the other speaker’s vocal tracts. A subsequent simulation produces the original sentence but with the vocal tract characteristics of the new speaker. The results also suggest that the time-varying modal coefficients may define common gestures across speakers even though their acoustic characteristics can be different.

Estimating the nonacoustic fluctuating velocity amplitude from Cranen and Boves’ vowel production data

In a dynamic mechanical model of the larynx and vocal tract [Barney et al., J. Acoust. Soc. Am. (in press)], the fluctuating velocity field was shown to have contributions from an acoustic disturbance and also from nonacoustic flow associated with vortex transport along the model tract. Nonacoustic velocity fluctuations, interacting with changes in the tract area function, can constitute significant sources of sound; in the vocal tract these could occur at the lips, teeth, etc. However, such additional sources of sound are not generally included in models of voiced speech production. Cranen and Boves [J. Acoust. Soc. Am. 77, 1543–1551 (1985)] obtained in-tract and radiated sound-pressure time histories from phonating subjects, from which they computed in-tract velocities. In this study the relative amplitudes of the acoustic and nonacoustic velocity fluctuations have also been estimated from their data. Though vocal tract area functions are not available, simple models of sound transmission and radiation allow an order-of-magnitude prediction of the contribution to the radiated sound associated with each part of the in-tract velocity field. Although these are preliminary investigations using a restricted data set, the evidence suggests that there is nonacoustic fluctuating flow within the vocal tract, capable of significant sound generation.

Estimation of the vocal tract area function from the impulse response at the lips

Estimation of the vocal tract area function from the impulse response at the lips

The estimation of vocal tract area function from speech signals : effects of the inner loss and wall impedance

The estimation of vocal tract area function from speech signals : effects of the inner loss and wall impedance

Parameterization of vocal tract area functions by empirical orthogonal modes

A set of ten vowel area functions, based on MRI measurements, has been parameterized by an “empirical orthogonal mode decomposition” which accurately represents each area function as the sum of the mean area function and proportional amounts of a series of orthogonal basis functions. The mean area function was found to possess a formant structure similar to that of a uniform tube (i.e., nearly equally spaced formants) suggesting that empirical orthogonal modes are perturbations on the mean (∼neutral) vowel shape much like past vocal tract analyses have considered perturbations on a uniform tube. The acoustic characteristics of the two most significant empirical orthogonal modes were examined, showing that both modes tend to increase the first formant as the modal amplitude coefficients are both increased from negative to positive values. However, the second formant was found to decrease in frequency for increasing values of the first modal coefficient and to increase for increasing values of the second mode coefficient. Next, a mapping between F1-F2 formant pairs and vocal tract area functions is proposed which is largely one-to-one but was initially limited by a constant vocal tract length. A possible method to include variable vocal tract length and higher ordered orthogonal modes in the mapping is given. The mode-to-formant mapping suggested the possibility of an inverse mapping to determine physiologically realistic area functions from a speech waveform and a simple example is presented. Finally, empirical orthogonal modes for a collection of ten vowels and eight consonants were derived and showed many similarities to those for the vowel-only case.

Simulation of sentence-level speech based on measured vocal tract area functions

An inventory of vocal tract area functions, acquired from magnetic resonance imaging experiments, is used as the basis for creating simulations of sentence-level speech. A first approach is a ‘‘brute-force’’ method in which a phonetic transcription of a sentence is time aligned with the recorded acoustic signal. Each element of the transcription is assigned one of the stored vocal tract area functions in the acquired inventory, thus serving as a ‘‘target’’ through which the time-varying area function should pass. This method has produced intelligible speech but maintains a machinelike quality. A second approach uses a parametric representation of the MRI-acquired area function inventory based on a principal components analysis to generate area functions not in the original inventory. The first three formants of each generated area function were computed and stored in a database, thus creating a mapping between formant frequencies and vocal tract parameters (area functions). Except for the edge of the formant space, the mapping between formants and area functions is one-to-one. This method produces connected speech with natural sounding vowels when only a microphone signal is used but requires additional signals to derive the consonantal contribution to the vocal tract shape. [Supported by NIH grant RO1 DC02532.]

Glottal source parameter estimation by comparison of measured signals with simulated signals

Oral flow, electroglottographic (EGG), and photoglottographic (PGG) signals were simultaneously digitized from a single subject phonating sustained vowels. Analogous simulated signals—oral flow, vocal fold contact area, and glottal area, respectively—were produced by a driven, kinematic glottal source model coupled to a wave-reflection-type vocal tract. The vocal tract area function used in the simulation was estimated from magnetic resonance imaging (MRI) data previously published for the subject. This essentially removed the vocal tract as a confounding factor in the analysis. The measured and simulated signals were compared and an attempt was made to adjust model parameters so that the least-squared differences between the signals were minimized. The model parameters, which included subglottal pressure, vocal fold adduction, vocal fold convergence, and fundamental frequency were observed to be estimates for the glottal configuration and control parameters used by the subject.

Acoustic measurement of vocal tract area function: An experimental study

The aim of this study was to develop a method for estimating the cross-sectional area function of the vocal tract by acoustic measurement. In this method, the area function is estimated using the transient response determined from the incident and reflected signals measured at the lips. The measurements are made with the vocal tract at one end of uniform tube that is excited at the other end by a speaker. With this method, it is extremely difficult to estimate the area function behind the narrow constriction, because the input and output signals are measured on the same side of the vocal tract. To improve estimation accuracy, a wide range and high energy level of frequency characteristics is needed for the input signal. The method for obtaining the desired input signal introduced here is based on the LMS error of the speaker output. To verify the estimation accuracy, an experiment was carried out using some cavities of known geometry. This measuring system shows good performance for the cavity whose ratio of minimum to maximum area is about 1:20. The estimated area functions of the human vocal tract were compared with midsagittal shapes simultaneously measured by a magnetic position sensing device.

A shape-based approach to vocal tract area function estimation

In default of a method for obtaining dynamic/time-varying 3-D vocal tract data, it remains the case that the midsagittal profile provides the best characterization of tract articulators. For this reason articulatory approaches to speech synthesis typically model the vocal tract as a series of concatenated tubes whose cross-sectional areas are related by some heuristic (the area function) to the midsagittal cross dimensions. But while areas alone are adequate for lower formants, accurate modeling of higher frequencies requires access to details of tract morphology. Previous work based on MRI volume parametrization from a single subject [Tiede et al., Proc. ETRW-SPM, 41–44 (1996)] has demonstrated the feasibility of recovering correlated cross-sectional shapes from the midsagittal profile alone. In the current study this approach is extended to analysis of four English and four Japanese subjects (five MRI-scanned vowels per subject). The results show that characteristic tract shapes for vowels can be recovered from a small and highly constrained set of parameters. Since area is easily recoverable from such shapes, this approach can be used to augment standard area function techniques.

Accurate recovery of articulator positions from acoustics: New conclusions based on human data

Vocal tract models are often used to study the problem of mapping from the acoustic transfer function to the vocal tract area function (inverse mapping). Unfortunately, results based on vocal tract models are strongly affected by the assumptions underlying the models. In this study, the mapping from acoustics (digitized speech samples) to articulation (measurements of the positions of receiver coils placed on the tongue, jaw, and lips) is examined using human data from a single speaker: Simultaneous acoustic and articulator measurements made for vowel-to-vowel transitions, /g/ closures, and transitions into and out of /g/ closures. Articulator positions were measured using an EMMA system to track coils placed on the lips, jaw, and tongue. Using these data, look-up tables were created that allow articulator positions to be estimated from acoustic signals. On a data set not used for making look-up tables, correlations between estimated and actual coil positions of around 94% and root-mean-squared errors around 2 mm are common for coils on the tongue. An error source evaluation shows that estimating articulator positions from quantized acoustics gives root-mean-squared errors that are typically less than 1 mm greater than the errors that would be obtained from quantizing the articulator positions themselves. This study agrees with and extends previous studies of human data by showing that for the data studied, speech acoustics can be used to accurately recover articulator positions.

Vocal tract area functions from magnetic resonance imaging

There have been considerable research efforts in the area of vocal tract modeling but there is still a small body of information regarding direct 3-D measurements of the vocal tract shape. The purpose of this study was to acquire, using magnetic resonance imaging (MRI), an inventory of speaker-specific, three-dimensional, vocal tract air space shapes that correspond to a particular set of vowels and consonants. A set of 18 shapes was obtained for one male subject who vocalized while being scanned for 12 vowels, 3 nasals, and 3 plosives. The 3-D shapes were analyzed to find the cross-sectional areas evaluated within planes always chosen to be perpendicular to the centerline extending from the glottis to the mouth to produce an "area function." This paper provides a speaker-specific catalogue of area functions for 18 vocal tract shapes. Comparisons of formant locations extracted from the natural (recorded) speech of the imaged subject and from simulations using the newly acquired area functions show reasonable similarity but suggest that the imaged vocal tract shapes may be somewhat centralized. Additionally, comparisons of the area functions reported in this study are compared with those from four previous studies and demonstrate general similarities in shape but also obvious differences that can be attributed to differences in imaging techniques, image processing methods, and anatomical differences of the imaged subjects.