Vocal Tract Model Research Articles

Estimation of area function for three-dimensional vocal tract using finite-element method

Numerous studies have obtained a vocal tract area function using data obtained from magnetic resonance imaging (MRI). However, this approach often results in increased error between formant frequencies derived by recorded voice and estimated frequencies derived by MRI. Two factors engender error: creating the vocal tract model and estimating the area function from the vocal tract model. Generally, the errors from those two factors are not considered independently. This study is intended to establish precise estimation of the area function automatically from a vocal tract model. Pressure contours and sound intensity in the vocal tract are obtained using the finite-element method (FEM); the vocal tract area function is obtained from them. The precision is assessed by comparing the formant frequencies that are derived using the FEM with our method. Results show that the vocal tract area function can be estimated precisely from three-dimensional MRI.

Acoustic demonstrations for education in speech science

Acoustic demonstrations are highly effective for education in speech science. We recently developed two educational tools for acoustics. The first educational tool is a set of physical models of the human vocal tract and related models. It contains cylinder and plate‐type models [T. Arai, J. Phonetic Soc. Jpn., 5(2), 31–38 (2001)], a sliding three‐tube model [T. Arai, Acoust. Sci. Technol., to be published], lung models, an artificial larynx, and head‐shaped models [T. Arai, Acoust. Sci. Technol., 27(2), 111–113 (2006)]. Each model has its own advantages and, if combined effectively, can produce a systematic and comprehensive education in speech production from the lungs to the head. The second educational tool is ‘‘Digital Pattern Playback (DPP)’’ [T. Arai et al., Acoust. Sci. Tech., to be published], which converts a spectrographic image of a speech signal back to sound by digital signal processing. A printed spectrogram on a sheet of paper can also be converted immediately after capturing the image from a camera, and we confirmed that this is more intuitive for learners than converting from an electronic image. [Work partially supported by JSPS.KAKENHI (17500603).]

Let children experience speech science

An education in acoustics is important for students of different ages: college, high school, middle school, and potentially even elementary school students, for whom such hands‐on models are the most important. Because speech communication is a basic human activity, even for children, we expect that even the youngest students to be interested in speech science. With this in mind, we were motivated to develop intuitive and effective models for teaching speech science to all types of students. Our physical models of the human vocal tract [T. Arai, J. Phonetic Soc. Jpn., 5(2), 31–38 (2001)] have recently been installed in an exhibition in a museum in Japan. Children can watch, listen to, and touch the models. Our recent head‐shaped and lung models [T. Arai, Acoust. Sci. Technol. 27(2), 111–113 (2006)] are also useful for educating students, because they are intuitive to produce a systematic and comprehensive education from the lungs to the head. Furthermore, we recently designed a sliding three‐tube (S3T) model that has a simple mechanism for producing several different vowels [T. Arai, Acoust. Sci. Technol (to be published)]. This model is being used for an activity at a science workshop for children. [Work partially supported by JSPS.KAKENHI (17500603).]

Finite-element method analysis of acoustic characteristics of the vocal tract with the nasal cavity during phonation of Japanese /a/

For this study, the transfer functions and active sound intensities of a vocal tract model with and without a nasal cavity were computed using a three-dimensional finite-element method (FEM). The models were based on vowel data obtained by magnetic resonance imaging (MRI) of the vocal tract with a nasal cavity during phonation of the Japanese /a/. The oral cavity was also coupled with the nasal cavity in a three-dimensional volume of radiation. Effects of wall impedance were also examined. Coupling of the nasal cavity to the oral cavity indicated the following aspects. Additional peaks appeared below 3 kHz for the lossless condition. However, they disappeared in the simulation for the soft wall condition. The sound energy circulation did not occur in the simulation for the soft wall condition. Regarding effects of the wall boundary condition on the spectral envelope, three-dimensional simulations confirmed the upward shift of lower formant frequencies. However, disagreement of the formant frequencies between simulated and real speech should be investigated further by adjusting the wall boundary condition to a more realistic one.

An MRI‐based time‐domain speech synthesis system

A speech synthesis system was developed based on Maedas method [S. Maeda, Speech Commun. 1, 199–229 (1982)], which simulates acoustic wave propagation in the vocal tract in the time domain. This system has a GUI interface that allows fine control of synthesis parameters and timing. In addition, the piriform fossae were included to the vocal tract model, resulting in antiresonances in speech spectra at the frequency region from 4 to 5 kHz. The system can produce all the Japanese phonemes using vocal tract area‐functions (VTAFs) extracted from 3‐D cine‐MRI obtained during production of VCV or CVCV sequences for a male speaker. The system can be used to synthesize Japanese sentences with high naturalness and intelligibility by concatenating segmental units and controlling the glottal source using the GUI interface. Since a time‐varying VTAF is obtained by interpolating between VTAFs, the dataset size of the system is significantly smaller than that of corpus‐based speech synthesizers. The speaker‐specific VTAFs and inclusion of the piriform fossae permit us to reproduce speaker‐specific spectral shapes, not only the lower formants but also higher frequency regions that contribute to the perception of speaker individualities. [Work supported by NICT, SCOPE‐R, and Grant‐in‐And for Scientific Research of Japan.]

Vocal-tract modeling: fractional elongation of segment lengths in a waveguide model with half-sample delays

Digital waveguide models are commonly used for simulating vocal-tract acoustics based on physiological data. In particular, waveguide models with half-sample delays are known to be well suited for speech production research. This paper presents enhancements to such a model, aimed at improved accuracy in mapping physiological vocal-tract data (shape and length of the airway) to waveguide parameters. The enhancements allow the length of the vocal tract to be continuously varied, thus enabling more realistic synthesis. This is achieved by smoothly varying the individual segment lengths of a piecewise-cylindrical representation of the airway, without altering the system sampling frequency. Fractional-delay filters are used for spatial interpolation of the digital waveguide model. The algorithms are validated by modeling the protrusion of lips, lowering of larynx and lengthening of intermediate segments for a static vowel shape

Singing synthesis with an evolved physical model

A two-dimensional physical model of the human vocal tract is described. Such a system promises increased realism and control in the synthesis of both speech and singing. However, the parameters describing the shape of the vocal tract while in use are not easily obtained, even using medical imaging techniques, so instead a genetic algorithm (GA) is applied to the model to find an appropriate configuration. Realistic sounds are produced by this method. Analysis of these, and the reliability of the technique (convergence properties) is provided

Computational aeroacoustics of flow through static model of human vocal tract

Computational aeroacoustics studies of flow through the human vocal tract, here modeled as a planar channel with an orifice, hence referred to as the glottis, were conducted using large-eddy simulation (LES). Comparisons were made between LES predictions and experimental wall pressure measurements and particle-imaging-velocimetry flow fields. Rigid models of both converging and diverging glottal passages, each featuring a 20-deg included angle and a minimal glottal width of 0.04 cm with transglottal pressure of 15 cm H2O, were studied. The compressible Navier-Stokes equations were accurately and efficiently integrated for the low Mach number flow through the use of an additive semi-implicit Runge-Kutta method and high-order compact finite-difference schemes for spatial discretization. Characteristic-based nonreflecting boundary conditions were used together with an exit zone in the context of a multiblock approach. Asymmetry of the flow due to a Coanda effect, and transition to turbulence were observed. An acoustic analogy based on the Ffowcs Williams-Hawkings equation was applied to decompose the near-field acoustic source into its monopole, dipole, and quadrupole contributions to assess glottal geometry effects on far-field sound. The results showed that dipole sources due to the unsteady forces exerted on the duct wall are dominant. [Work funded by NIH Grant RO1 DC03577.]

An In-vitro Test Set-Up for Evaluation of a Voice-Producing Element Under Physiologic Acoustic Conditions

To improve the voice quality of laryngectomised patients, a voice-producing element has been developed. Prior to in vivo testing we constructed and validated an in-vitro test set-up, consisting of a physical model of the subglottal tract and three physical models of the vocal tract, for the vowels /a/, /i/ and /u/ to evaluate the voice-producing element under physiologic acoustic conditions. To meet acoustic conditions described in the literature, we determined the appropriate dimensions of these physical models, using a numerical model of the pressure perturbation in rigid tubes. The numerical model showed that an acoustic equivalent of the subglottal tract could be obtained with a three-tube system and an end impedance. Vocal tract models could be constructed using two- and four-resonator tubes. The physical models were built and successfully validated according to the human acoustic properties. The developed in-vitro set-up can now be applied to test voice-producing elements or vocal fold models under physiologic acoustic conditions.

Vocal Tract Area Function for Vowels Using Three-Dimensional Magnetic Resonance Imaging. A Preliminary Study

To assess whether magnetic resonance imaging (MRI) allows the vocal tract (VT) area function to be determined for a normal male speaker. VT shapes were acquired using MRI during sustained production of French points vowels: /i/, /a/, /u/. Cross-sectional areas were measured from a series of planes spaced at intervals of 1 cm along the length of the VT and were used as input in a previously described VT model to simulate the vowels. The first three formant frequencies, F1, F2, and F3, computed from the MRI-measured VT model were compared with subject's natural formant frequencies. Including piriform sinuses, calculated formants differed from measured formants F1, F2, and F3, respectively, for /i/ by -3.5%, +7.7%, and +27.5%; for /a/ by +11% +19.5%, and -4.3%; and for /u/ by +.9%, +23.4%, and +9.6%. Excluding piriform sinuses, calculated formants differed from measured formants F1, F2, and F3, respectively, for /i/ by -3.5%, +12%, and +28%, and for /u/ by +10.1%, +26.8%, and +13.7% The piriform sinuses were not visualized for /a/ on MRI. MRI is a noninvasive technique that allows VT imaging and determination of VT area function for a normal male speaker. Several possible sources of discrepancies are as follows: variability of the articulation, difficulties in assessment of VT wall boundaries, role of the piriform sinuses, and VT length.

Methods for formant extraction in speech of patients after total laryngectomy

The paper shows the methods and its application for voice analysis suited to the group of subjects after total laryngectomy surgery. Our software was developed to evaluate and enhance laryngectomized patients’ rehabilitation process. The power spectral density imaging and formant frequencies extraction methods were adopted. The model of vocal tract was based on statistical, autoregressive process of speech production. The transversal filter and adaptive algorithm were implemented to estimate the transfer function of resonance cavities. The research is concerned with measurements of vowel articulation parameters, especially F1 and F2 formant frequencies. The significant difference of pathological and normal voice in vowel space separation has been presented. The authors found that formants in pseudowhisper speech are more pronounced while articulating vowel after consonant than for sustained vowel.

Robust Glottal Source Estimation Based on Joint Source-Filter Model Optimization

This paper describes a robust glottal source estimation method based on a joint source-filter separation technique. In this method, the Liljencrants–Fant (LF) model, which models the glottal flow derivative, is integrated into a time-varying ARX speech production model. These two models are estimated in a joint optimization procedure, in which a Kalman filtering process is embedded for adaptively identifying the vocal tract parameters. Since the formulated joint estimation problem is a multiparameter nonlinear optimization procedure, we separate the optimization procedure into two passes. The first pass initializes the glottal source and vocal tract models by solving a quasi-convex approximate optimization problem. Having robust initial values, the joint estimation procedure determines the accuracy of model estimation implemented with a trust-region descent optimization algorithm. Experiments with synthetic and real voice signals show that the proposed method is a robust glottal source parameter estimation method with a high degree of accuracy.

An investigation of bimodal jet trajectory in flow through scaled models of the human vocal tract

Pulsatile two-dimensional flow through static divergent models of the human vocal folds is investigated. Although the motivation for this study is speech production, the results are generally applicable to a variety of engineering flows involving pulsatile flow through diffusers. Model glottal divergence angles of 10, 20, and 40° represent various geometries encountered in one phonation cycle. Frequency and amplitude of the flow oscillations are scaled with physiological Reynolds and Strouhal numbers typical of human phonation. Glottal velocity trajectories are measured along the anterior–posterior midline by using phase-averaged particle image velocimetry to acquire 1,000 realizations at ten discrete instances in the phonation cycle. The angular deflection of the glottal jet from the streamwise direction (symmetric configuration) is quantified for each realization. A bimodal flow configuration is observed for divergence angles of 10 and 20°, with the flow eventually skewing and attaching to the vocal fold walls. The deflection of the flow toward the vocal fold walls occurs when the forcing function reaches maximum velocity and zero acceleration. For a divergence angle of 40°, the flow never attaches to the vocal fold walls; however, there is increased variability in the glottal jet after the forcing function reaches maximum velocity and zero acceleration. The variation in the jet trajectory as a function of divergence angle is explained by performance maps of diffuser flow regimes. The smaller angle cases are in the unstable transitory stall regime while the 40° divergent case is in the fully developed two-dimensional stall regime. Very small geometric variations in model size and surface finish significantly affect the flow behavior. The bimodal, or flip-flopping, glottal jet behavior is expected to influence the dipole contribution to sound production.

Demonstrations for education in acoustics in Japan

Many demonstrations for education in acoustics have been developed in Japan as well as outside the country. Since 1997, the Technical Committee on Education in Acoustics of the Acoustical Society of Japan has been investigating and discussing education in acoustics in Japan. In this review, some of the educational tools and demonstrations in acoustics are introduced. They are all designed to help us visualize and hear different phenomena and to understand abstract theories in a more intuitive way. The work that has been carried out includes some exciting demonstrations in acoustics by the high-school physics teachers' ''Stray Cats Group,'' some visual and aural demonstrations for architectural acoustics, a technical course called ''Technical Listening Training,'' a WWW-based training system, and physical models of the human vocal tract.

Phonation thresholds as a function of laryngeal size in a two-mass model of the vocal folds

This letter analyzes the oscillation onset-offset conditions of the vocal folds as a function of laryngeal size. A version of the two-mass model of the vocal folds is used, coupled to a two-tube approximation of the vocal tract in configuration for the vowel /a/. The standard male configurations of the laryngeal and vocal tract models are used as reference, and their dimensions are scaled using a single factor. Simulations of the vocal fold oscillation and oral output are produced for varying values of the scaling factor. The results show that the oscillation threshold conditions become more restricted for smaller laryngeal sizes, such as those appropriate for females and children.

An experimental investigation of velocity fields in divergent glottal models of the human vocal tract

In speech, sound production arises from fluid-structure interactions within the larynx as well as viscous flow phenomena that is most likely to occur during the divergent orientation of the vocal folds. Of particular interest are the flow mechanisms that influence the location of flow separation points on the vocal folds walls. Physiologically scaled pulsatile flow fields in 7.5 times real size static divergent glottal models were investigated. Three divergence angles were investigated using phase-averaged particle image velocimetry (PIV). The pulsatile glottal jet exhibited a bi-modal stability toward both glottal walls, although there was a significant amount of variance in the angle the jet deflected from the midline. The attachment of the Coanda effect to the glottal model walls occurred when the pulsatile velocity was a maximum, and the acceleration of the waveform was zero. The location of the separation and reattachment points of the flow from the glottal models was a function of the velocity waveform and divergence angle. Acoustic analogies show that a dipole sound source contribution arising from the fluid interaction (Coanda jet) with the vocal fold walls is expected. [Work funded by NIH Grant RO1 DC03577.]

Visualizing vowel‐production mechanism using simple educational tools

To develop intuitive and effective methods for educating Acoustics to students of different ages and from varied backgrounds, Arai [J. Phonetic Soc. Jpn. 5, 31–38, (2001)] replicated Chiba and Kajiyama’s physical models of the human vocal tract as educational tools and verified that the physical models and sound sources, such as an artificial larynx, yield a simple but powerful demonstration of vowel production in the classroom. We have also started exhibiting our models at the Science Museum ‘‘Ru‐Ku‐Ru’’ in Shizuoka City, Japan. We further extended our model to a lung model as well as several head‐shaped models with visible vocal tract to demonstrate the total vowel‐production mechanism from phonation to articulation. The lung model imitates the human respiratory system with a diaphragm. In the head‐shaped model, the midsaggital cross section is visible from the outside. To adjust the degree of nasopharyngeal coupling, the velum may be rotated. Another head‐shaped model with the manipulable tongue position was also developed. Two test results were compared before and after using these physical models, and the educational effectiveness of the models was confirmed. The homepage of the vocal‐tract models is available at http://www.splab.ee.sophia.ac.jp/Vocal—Tract—Model/index‐e.htm. [Work supported by KAKENHI (17500603).]

Robust processing techniques for voice conversion

Differences in speaker characteristics, recording conditions, and signal processing algorithms affect output quality in voice conversion systems. This study focuses on formulating robust techniques for a codebook mapping based voice conversion algorithm. Three different methods are used to improve voice conversion performance: confidence measures, pre-emphasis, and spectral equalization. Analysis is performed for each method and the implementation details are discussed. The first method employs confidence measures in the training stage to eliminate problematic pairs of source and target speech units that might result from possible misalignments, speaking style differences or pronunciation variations. Four confidence measures are developed based on the spectral distance, fundamental frequency ( f 0) distance, energy distance, and duration distance between the source and target speech units. The second method focuses on the importance of pre-emphasis in line-spectral frequency (LSF) based vocal tract modeling and transformation. The last method, spectral equalization, is aimed at reducing the differences in the source and target long-term spectra when the source and target recording conditions are significantly different. The voice conversion algorithm that employs the proposed techniques is compared with the baseline voice conversion algorithm with objective tests as well as three subjective listening tests. First, similarity to the target voice is evaluated in a subjective listening test and it is shown that the proposed algorithm improves similarity to the target voice by 23.0%. An ABX test is performed and the proposed algorithm is preferred over the baseline algorithm by 76.4%. In the third test, the two algorithms are compared in terms of the subjective quality of the voice conversion output. The proposed algorithm improves the subjective output quality by 46.8% in terms of mean opinion score (MOS).

A modeling investigation of articulatory variability and acoustic stability during American English /r/ production

This paper investigates the functional relationship between articulatory variability and stability of acoustic cues during American English /r/ production. The analysis of articulatory movement data on seven subjects shows that the extent of intrasubject articulatory variability along any given articulatory direction is strongly and inversely related to a measure of acoustic stability (the extent of acoustic variation that displacing the articulators in this direction would produce). The presence and direction of this relationship is consistent with a speech motor control mechanism that uses a third formant frequency (F3) target; i.e., the final articulatory variability is lower for those articulatory directions most relevant to determining the F3 value. In contrast, no consistent relationship across speakers and phonetic contexts was found between hypothesized vocal-tract target variables and articulatory variability. Furthermore, simulations of two speakers' productions using the DIVA model of speech production, in conjunction with a novel speaker-specific vocal-tract model derived from magnetic resonance imaging data, mimic the observed range of articulatory gestures for each subject, while exhibiting the same articulatory/acoustic relations as those observed experimentally. Overall these results provide evidence for a common control scheme that utilizes an acoustic, rather than articulatory, target specification for American English /r/.

An oscillator-plus-noise model for speech synthesis

The autonomous oscillator model for speech synthesis is augmented by a non-linear predictor to re-generate the modulated noise-like signal component of speech signals. The resulting ‘oscillator-plus-noise’ model in combination with vocal tract modeling by linear prediction is able to re-generate the spectral content of stationary wide-band vowel signals with high fidelity. For adequate modeling of mixed-excitation speech signals (such as voiced fricatives), the model is extended by a second linear prediction path for the independent spectral shaping of the noise-like component. With one and the same model, not only sustained voiced and mixed-excitation phonemes, but also stationary unvoiced sounds can be re-generated faithfully.