Vocal Tract Model Research Articles

Comparison of waveform estimated using inverse filtering with direct measurement of the volume flow at the glottal exit

Inverse filtering is a signal processing technique used to estimate the glottal waveform from the speech signal. Although this technique is regularly used in research studies, it has never been formally validated because of the difficulties in obtaining direct measurements of the glottal airflow. The objective of this study is a first step towards validating this technique by comparing its estimated glottal waveform with flow measurements taken simultaneously at the glottal exit. The setup is based on synthetic vocal folds connected to a vocal tract model. Direct measurements of the volume flow at the glottal exit are taken using time-resolved tomographic particle image velocimetry. A circumferentially vented pneumotachograph (i.e., Rothenberg) mask is connected to the vocal tract and used with the Glottal Enterprise system to calculate the glottal waveform using inverse filtering. Effect of varying subglottal pressures and effect of near-field (i.e., the minimal gap between the false vocal folds) and far-field (i.e., oral opening) constrictions were also investigated. Results show that overall using inverse filtering gives a good approximation of the glottal flow waveform, but its accuracy can change depending on the constrictions in the vocal tract. The clinical implication of these findings will be further discussed.

Determination of the vocal tract model order in iterative adaptive inverse filtering

This paper is part of a special issue on Music Technology. Accurate estimation of glottal wave is an important method for singing voice analysis via voicing patterns. In this paper, an optimising method is proposed for automatic determination of vocal tract linear prediction analysis order that follows the specific situation of different voicing scenes based on Iterative Adaptive Inverse Filtering (IAIF) aiming at getting a more accurate glottal wave from speech or singing voice signal in a non-invasive way. In comparison with existing methods that use fixed experience order, the proposed method can achieve up to 8.41% improvement in correlation coefficient with the real glottal wave.

Analyzing the bass whistling method using a vocal tract model

Analyzing the bass whistling method using a vocal tract model

ACCURACY OF DETERMINATION OF LINEAR SPECTRAL FREQUENCIES

The article describes the basic idea of coding a speech signal by the method of linear prediction (Linear Predictive Coding – LPC), which consists in the fact that instead of the parameters of the speech signal, the encoded parameters of a certain filter are transmitted over the communication line, which is, in a sense, an equivalent of the human vocal tract, as well as parameters of the excitation signal of this filter: tone or noise. The essence of the parameters of the synthesizing filter – linear prediction coefficients (LP), calculated in the process of frame-by-frame adaptive filtering – is disclosed. The main advantages of these coefficients are stated, which consists in the ability to completely describe the state of the predictor filter, as well as the main disadvantages obtained as a result of numerous studies that prevent the direct transmission of LPCs over the communication channel due to their sensitivity to quantization errors. The necessity of searching for mathematically equivalent parameters f the reducing filter is substantiated. Alternative parameters of the LSP representation of the vocal tract model, called Linear Spectral Frequencies (LSP), are proposed, and are most often used in low-speed speech codecs at the present time. The main difficulty of the task of calculating the LSP directly from the LSP in real time is described, and it is also investigated that most of the analyzer’s processor time, as a rule, is spent on this task. It is proved that despite the high level of development of digital processors, the problem of calculating the parameters of the vocal tract model in real time remains one of the main difficulties in speech coding. That is why methods that reduce the complexity of this procedure are currently of significant interest. A new method for calculating the LSP and an algorithm based on it are proposed. Its main advantages over the existing ones are described. An assessment of the possible improvement in the quality of speech processing was made, due to the use of the developed method for calculating the LSP.

Estimating Underlying Articulatory Targets of Thai Vowels by Using Deep Learning Based on Generating Synthetic Samples From a 3D Vocal Tract Model and Data Augmentation

Representation learning is one of the fundamental issues in modeling articulatory-based speech synthesis using target-driven models. This paper proposes a computational strategy for learning underlying articulatory targets from a 3D articulatory speech synthesis model using a bi-directional long short-term memory recurrent neural network based on a small set of representative seed samples. Using a seeding set from VocalTractLab, a larger training set was generated that provided richer contextual variations for the model to learn. The deep learning model for acoustic-to-target mapping was then trained to model the inverse relation of the articulation process. This method allows the trained model to map the given acoustic data onto the articulatory target parameters which can then be used to identify the distribution based on linguistic contexts. The model was evaluated based on its effectiveness in mapping acoustics to articulation, and the perceptual accuracy of speech reproduced from the articulation estimated from the recorded speech by native Thai speakers. The model achieved more than 80% phoneme classification accuracy in the listening test conducted with 25 native Thai speakers. The results indicate that the model can accurately imitate speech with a high degree of phonemic precision.

Investigation of resonance strategies of high pitch singing sopranos using dynamic three-dimensional magnetic resonance imaging.

Resonance-strategies with respect to vocal registers, i.e., frequency-ranges of uniform, demarcated voice quality, for the highest part of the female voice are still not completely understood. The first and second vocal tract resonances usually determine vowels. If the fundamental frequency exceeds the vowel-shaping resonance frequencies of speech, vocal tract resonances are tuned to voice source partials. It has not yet been clarified if such tuning is applicable for the entire voice-range, particularly for the top pitches. We investigated professional sopranos who regularly sing pitches above C6 (1047 Hz). Dynamic three-dimensional (3D) magnetic resonance imaging was used to calculate resonances for pitches from C5 (523 Hz) to C7 (2093 Hz) with different vowel configurations ([a:], [i:], [u:]), and different contexts (scales or octave jumps). A spectral analysis and an acoustic analysis of 3D-printed vocal tract models were conducted. The results suggest that there is no exclusive register-defining resonance-strategy. The intersection of fundamental frequency and first vocal tract resonance was not found to necessarily indicate a register shift. The articulators and the vocal tract resonances were either kept without significant adjustments, or the fR1:fo-tuning, wherein the first vocal tract resonance enhances the fundamental frequency, was applied until F6 (1396 Hz). An fR2:fo-tuning was not observed.

Method for Fabricating Transparent Patient-Specific Vocal Tract Replicas.

Transparent, patient-specific vocal tract replicas are helpful in research and educational endeavors but challenging to procure. An accessible method for fabricating these models, improving on previously suggested processes, would make them more widely available. Detailed instructions for fabricating a transparent, patient-specific vocal tract model were addressed. The broad steps were (1) digitally reconstructing (patient-specific) vocal tract geometry, (2) producing a vocal tract mold (using methods such as three-dimensional [3D] printing), and (3) casting transparent material (such as silicone) around the vocal tract mold and removing the mold. The cavities remaining within the cast represented the exact geometry of the vocal tract. A combination of 3D printing and silicone casting can produce useful vocal tract replicas. Several simple changes to previous methods can improve consistency and reduce the labor and cost of production. Limitations and potential modifications to expand the applications of this method are discussed.

Classification of Parkinson disease based on analysis and synthesis of voice signal

The most important application of voice profiling is pathological voice detection. Parkinson's disease is a chronic neurological degenerative disease affecting the central nervous system responsible for essentially progressive evolution movement disorders. 70% to 90% of Parkinson’s disease (PD) patients show an affected voice. This paper proposes a methodology for PD based on acoustic, glottal, physical, and electrical parameters. The results show that the acoustic parameter is more important in the case of Parkinson’s disease as compared to glottal and physical parameters. The authors achieved 97.2% accuracy to differentiate Parkinson and healthy voice using jitter to pitch ratio proposed algorithm. The Authors also proposed an algorithm of poles calculation of the vocal tract to find formants of the vocal tract. Further, formants are used for finding the transfer function of vocal tract filter. In the end, the authors suggested parameters of the electrical vocal tract model are also changed in the case of PD voices.

Classification of Parkinson disease based on analysis and synthesis of voice signal

The most important application of voice profiling is pathological voice detection. Parkinson's disease is a chronic neurological degenerative disease affecting the central nervous system responsible for essentially progressive evolution movement disorders. 70% to 90% of Parkinson’s disease (PD) patients show an affected voice. This paper proposes a methodology for PD based on acoustic, glottal, physical, and electrical parameters. The results show that the acoustic parameter is more important in the case of Parkinson’s disease as compared to glottal and physical parameters. The authors achieved 97.2% accuracy to differentiate Parkinson and healthy voice using jitter to pitch ratio proposed algorithm. The Authors also proposed an algorithm of poles calculation of the vocal tract to find formants of the vocal tract. Further, formants are used for finding the transfer function of vocal tract filter. In the end, the authors suggested parameters of the electrical vocal tract model are also changed in the case of PD voices.

Classification of Parkinson Disease Based on Analysis and Synthesis of Voice Signal

The most important application of voice profiling is pathological voice detection. Parkinson's disease is a chronic neurological degenerative disease affecting the central nervous system responsible for essentially progressive evolution movement disorders. 70% to 90% of Parkinson’s disease (PD) patients show an affected voice. This paper proposes a methodology for PD based on acoustic, glottal, physical, and electrical parameters. The results show that the acoustic parameter is more important in the case of Parkinson’s disease as compared to glottal and physical parameters. The authors achieved 97.2% accuracy to differentiate Parkinson and healthy voice using jitter to pitch ratio proposed algorithm. The Authors also proposed an algorithm of poles calculation of the vocal tract to find formants of the vocal tract. Further, formants are used for finding the transfer function of vocal tract filter. In the end, the authors suggested parameters of the electrical vocal tract model are also changed in the case of PD voices.

Bass whistling reproduction using a vocal tract model

Bass whistling reproduction using a vocal tract model

Effects of the piriform fossae, transvelar acoustic coupling, and laryngeal wall vibration on the naturalness of articulatory speech synthesis

Acoustic models of the vocal tract for articulatory speech synthesis often neglect a range of acoustic effects that are known to exist in the human vocal tract. Here we extended a basic acoustic vocal tract model by three features: the piriform fossae, transvelar acoustic coupling of the oral and nasal cavities, and sound radiation from the skin of the neck. The main goal was to find out how these features affect the naturalness of the synthesized speech. To this end, ten German words were synthesized with different combinations of the additional features, and listeners compared the naturalness of these stimuli. Surprisingly, all three features reduced the perceived naturalness, although they should make the synthesis more realistic. A closer analysis revealed that all new features emphasized the low frequencies compared to the high frequencies of the synthetic speech, leading to slightly more muffled speech with the used glottal excitation. An additional perception experiment with synthetic stimuli with a slightly more tense voice revealed no perceptual preference for the synthesis with or without the piriform fossae. These results indicate that the examined features play a minor role for the naturalness of articulatory synthesis compared to the voice source characteristics.

Aeroacoustic differences between the Japanese fricatives [ɕ] and [ç

To elucidate the linguistic similarity between the alveolo-palatal sibilant [ɕ] and palatal non-sibilant [ç] in Japanese, the aeroacoustic differences between the two consonants were explored via experimentation with participants and analysis using simplified vocal tract models. The real-time magnetic resonance imaging (rtMRI) observations of articulatory movements demonstrated that some speakers use a nearly identical place of articulation for /si/ [ɕi] and /hi/ [çi]. Simplified vocal tract models were then constructed based on the data captured by static MRI, and the model-generated synthetic sounds were compared with speaker data producing [ɕ] and [ç]. Speaker data demonstrated that the amplitude of the broadband noise of [ç] was weaker than that of [ɕ]; the characteristic peak amplitude at approximately 4 kHz was greater in [ç] than in [ɕ], although the mid-sagittal vocal tract profiles were nearly identical for three of ten subjects in the rtMRI observation. These acoustic differences were reproduced by the proposed models, with differences in the width of the coronal plane constriction and the flow rate. The results suggest the need to include constriction width and flow rate as parameters for articulatory phonetic descriptions of speech sounds.

On the Speech Properties and Feature Extraction Methods in Speech Emotion Recognition.

Many speech emotion recognition systems have been designed using different features and classification methods. Still, there is a lack of knowledge and reasoning regarding the underlying speech characteristics and processing, i.e., how basic characteristics, methods, and settings affect the accuracy, to what extent, etc. This study is to extend physical perspective on speech emotion recognition by analyzing basic speech characteristics and modeling methods, e.g., time characteristics (segmentation, window types, and classification regions—lengths and overlaps), frequency ranges, frequency scales, processing of whole speech (spectrograms), vocal tract (filter banks, linear prediction coefficient (LPC) modeling), and excitation (inverse LPC filtering) signals, magnitude and phase manipulations, cepstral features, etc. In the evaluation phase the state-of-the-art classification method and rigorous statistical tests were applied, namely N-fold cross validation, paired t-test, rank, and Pearson correlations. The results revealed several settings in a 75% accuracy range (seven emotions). The most successful methods were based on vocal tract features using psychoacoustic filter banks covering the 0–8 kHz frequency range. Well scoring are also spectrograms carrying vocal tract and excitation information. It was found that even basic processing like pre-emphasis, segmentation, magnitude modifications, etc., can dramatically affect the results. Most findings are robust by exhibiting strong correlations across tested databases.

3D-FV-FE Aeroacoustic Larynx Model for Investigation of Functional Based Voice Disorders.

For the clinical analysis of underlying mechanisms of voice disorders, we developed a numerical aeroacoustic larynx model, called simVoice, that mimics commonly observed functional laryngeal disorders as glottal insufficiency and vibrational left-right asymmetries. The model is a combination of the Finite Volume (FV) CFD solver Star-CCM+ and the Finite Element (FE) aeroacoustic solver CFS++. simVoice models turbulence using Large Eddy Simulations (LES) and the acoustic wave propagation with the perturbed convective wave equation (PCWE). Its geometry corresponds to a simplified larynx and a vocal tract model representing the vowel /a/. The oscillations of the vocal folds are externally driven. In total, 10 configurations with different degrees of functional-based disorders were simulated and analyzed. The energy transfer between the glottal airflow and the vocal folds decreases with an increasing glottal insufficiency and potentially reflects the higher effort during speech for patients being concerned. This loss of energy transfer may also have an essential influence on the quality of the sound signal as expressed by decreasing sound pressure level (SPL), Cepstral Peak Prominence (CPP), and Vocal Efficiency (VE). Asymmetry in the vocal fold oscillations also reduces the quality of the sound signal. However, simVoice confirmed previous clinical and experimental observations that a high level of glottal insufficiency worsens the acoustic signal quality more than oscillatory left-right asymmetry. Both symptoms in combination will further reduce the quality of the sound signal. In summary, simVoice allows for detailed analysis of the origins of disordered voice production and hence fosters the further understanding of laryngeal physiology, including occurring dependencies. A current walltime of 10 h/cycle is, with a prospective increase in computing power, auspicious for a future clinical use of simVoice.

Automatic detection of pharyngeal fricatives in cleft palate speech using acoustic features based on the vocal tract area spectrum

The pharyngeal fricative is a typical compensatory articulation disorder in cleft palate speech. It is produced by retracting the root of the tongue to the posterior pharyngeal wall to substitute for the fricatives and affricates produced in the oral cavity. People who use the pharyngeal fricative have difficulties in daily communication. Research on automatic pharyngeal fricative detection can provide aids in diagnosis for speech-language pathologists and clinical doctors. This work proposes a vocal tract area spectrum (VTAS) to represent a vocal tract model using time-varying cascaded pipes. Four acoustic features based on the VTAS (the centroid and spread (CS), peak linear deviation (PLD), relative-normal entropy (RNE), mean of the ratios’ statistics (MRS)) are proposed to evaluate the differences between pharyngeal fricatives and normal speech. The CS feature is proposed to evaluate the overall shape of the vocal tract to detect whether there are abnormal gestures or movements of the articulators in speech production. The PLD and RNE features focus on the variation and complexity of each vocal tube's area during the whole pronunciation process. The MRS feature is proposed to describe the continuity of the vocal tract. To evaluate the effectiveness of these four features, pharyngeal fricative detection experiments are conducted using a pharyngeal fricative dataset. This dataset contains 1246 speech samples spoken by 50 cleft palate patients and 50 normal speakers, covering all types of initial consonants in which the pharyngeal fricative usually occurs. The detection accuracy of the pharyngeal fricative using the CS, PLD, RNE and MRS feature ranges from 80.66% to 90.21%. When using the proposed CS+PLD+RNE+MRS feature, an accuracy of 95.18% can be achieved on the pharyngeal fricative dataset.

Goal-Directed Exploration for Learning Vowels and Syllables: A Computational Model of Speech Acquisition

Infants learn to speak rapidly during their first years of life, gradually improving from simple vowel-like sounds to larger consonant-vowel complexes. Learning to control their vocal tract in order to produce meaningful speech sounds is a complex process which requires to learn the relationship between motor and sensory processes. In this paper, a computational framework is proposed that models the problem of learning articulatory control for a physiologically plausible 3-D vocal tract model using a developmentally-inspired approach. The system babbles and explores efficiently in a low-dimensional space of goals that are relevant to the learner in its synthetic environment. The learning process is goal-directed and self-organized, and yields an inverse model of the mapping between sensory space and motor commands. This study provides a unified framework that can be used for learning static as well as dynamic motor representations. The successful learning of vowel and syllable sounds as well as the benefit of active and adaptive learning strategies are demonstrated. Categorical perception is found in the acquired models, suggesting that the framework has the potential to replicate phenomena of human speech acquisition.

Vocal Tract Parameter Estimation and Modeling with Application in Diagnosis of Voice Disorders

Vocal Tract modelling plays an important role in acoustic phonetics that gives a speech language therapist (SLP) information about the shape of the vocal tract. Vocal tract transfer function estimation gives important insights in diagnosing voice disorders. This paper presents a survey of vocal tract modelling methods employed in literature. Vocal tract parameter estimation using Vagmi and Praat are detailed in this paper, followed by an investigation into a method to measure the resonance of the vocal tract using frequency sweeps at the lips of the human subject. The design and implementation of the vocal tract resonance measurement device has been tested and has potential for application in diagnosis of voice disorders.

Finite-element modeling of vocal fold self-oscillations in interaction with vocal tract: Comparison of incompressible and compressible flow model

Finite-element modeling of self-sustained vocal fold oscillations during voice production has mostly considered the air as incompressible, due to numerical complexity. This study overcomes this limitation and studies the influence of air compressibility on phonatory pressures, flow and vocal fold vibratory characteristics. A two-dimensional finite-element model is used, which incorporates layered vocal fold structure, vocal fold collisions, large deformations of the vocal fold tissue, morphing the fluid mesh according to the vocal fold motion by the arbitrary Lagrangian-Eulerian approach and vocal tract model of Czech vowel [i:] based on data from magnetic resonance images. Unsteady viscous compressible or incompressible airflow is described by the Navier-Stokes equations. An explicit coupling scheme with separated solvers for structure and fluid domain was used for modeling the fluid-structure-acoustic interaction. Results of the simulations show clear differences in the glottal flow and vocal fold vibration waveforms between the incompressible and compressible fluid flow. These results provide the evidence on the existence of the coupling between the vocal tract acoustics and the glottal flow (Level 1 interactions), as well as between the vocal tract acoustics and the vocal fold vibrations (Level 2 interactions).

Vocal tract model for speech production of [s] considering incisor angle

Vocal tract model for speech production of [s] considering incisor angle