Vocal Tract Filter Research Articles

Symmetric and asymmetric Gaussian weighted linear prediction for voice inverse filtering

Weighted linear prediction (WLP) has demonstrated its significance in voice inverse filtering, contributing to enhanced methods for estimating both the vocal tract filter and the glottal source. WLP provides a mechanism to mitigate the effect on the linear prediction model of voice samples that affects the vocal tract filter estimation, particularly those samples around glottal closure instants (GCIs). This article studies the Gaussian weighted linear prediction (GLP) strategy, which employs a Gaussian attenuation window centered at the GCIs to reduce its contribution in the WLP analysis. In this study, the Gaussian attenuation is revisited and a parameterization of the window that adjusts to the typical variability in voice periodicity is introduced. In addition, an asymmetric Gaussian window is proposed to diminish the relevance of voice samples preceding GCIs on the WLP model, thus providing a quasi closed phase inverse filtering method. Characterization of symmetric and asymmetric GLP methods for glottal source estimation is addressed based on synthetic and natural phonation data, resulting in a set of optimal parameters for the Gaussian attenuation windows. The results show that the proposed asymmetric attenuation improves voice inverse filtering with respect to the symmetric GLP method. Comparisons with other state-of-the-art techniques suggest that the proposed GLP approaches are competitive, falling slightly short in performance only when contrasted with the well-known quasi closed inverse filtering analysis. The simplicity of implementing the attenuation windows, coupled with their robust performance, positions the proposed GLP methods as two attractive and straightforward voice inverse filtering techniques for practical application.

Characterization of Source-Filter Interactions in Vocal Vibrato Using a Neck-Surface Vibration Sensor: A Pilot Study.

Vocal vibrato is a singing technique that involves periodic modulation of fundamental frequency (fo) and intensity. The physiological sources of modulation within the speech mechanism and the interactions between the laryngeal source and vocal tract filter in vibrato are not fully understood. Therefore, the purpose of this study was to determine if differences in the rate and extent of fo and intensity modulation could be captured using simultaneously recorded signals from a neck-surface vibration sensor and a microphone, which represent features of the source before and after supraglottal vocal tract filtering. Nine classically-trained singers produced sustained vowels with vibrato while simultaneous signals were recorded using a vibration sensor and a microphone. Acoustical analyses were performed to measure the rate and extent of fo and intensity modulation for each trial. Paired-samples sign tests were used to analyze differences between the rate and extent of fo and intensity modulation in the vibration sensor and microphone signals. The rate and extent of fo modulation and the extent of intensity modulation were equivalent in the vibration sensor and microphone signals, but the rate of intensity modulation was significantly higher in the microphone signal than in the vibration sensor signal. Larger differences in the rate of intensity modulation were seen with vowels that typically have smaller differences between the first and second formant frequencies. This study demonstrated that the rate of intensity modulation at the source prior to supraglottal vocal tract filtering, as measured in neck-surface vibration sensor signals, was lower than the rate of intensity modulation after supraglottal vocal tract filtering, as measured in microphone signals. The difference in rate varied based on the vowel. These findings provide further support of the resonance-harmonics interaction in vocal vibrato. Further investigation is warranted to determine if differences in the physiological source(s) of vibrato account for inconsistent relationships between the extent of intensity modulation in neck-surface vibration sensor and microphone signals.

Autoregressive Based Vocal Tract Shape Modelling of Vowels in Speech Processing

In the present research article, which concentrates on the estimation of the vocal tract shape for the five vowels of Indian English in the current growing interest in the field of speech processing, with practical limitation of the data that can be collected and analysed. Speakers without the degree of control on articulators cannot produce the desired data. Here, we aim to design an approximate phonatory model by improving the present approach for acoustic calculation with a fully aerodynamic simulation, explicitly accounting for the propagation of sound along the tract, generating patterns of movement of simulated vocal tract articulators, and specifying the temporal relations among dynamically defined gestures that lead to a time-varying vocal tract filter function, and an acoustic waveform. The principle of phonetic distinctiveness and formants frequency spread is one of the crucial parameters that are measured and compared on intra, inter-speaker in actual speaker identification & forensic applications.

ACOUSTICALLY-DRIVEN PHONEME REMOVAL THAT PRESERVES VOCAL AFFECT CUES.

In this paper, we propose a method for removing linguistic information from speech for the purpose of isolating paralinguistic indicators of affect. The immediate utility of this method lies in clinical tests of sensitivity to vocal affect that are not confounded by language, which is impaired in a variety of clinical populations. The method is based on simultaneous recordings of speech audio and electroglotto-graphic (EGG) signals. The speech audio signal is used to estimate the average vocal tract filter response and amplitude envelop. The EGG signal supplies a direct correlate of voice source activity that is mostly independent of phonetic articulation. These signals are used to create a third signal designed to capture as much paralinguistic information from the vocal production system as possible-maximizing the retention of bioacoustic cues to affect-while eliminating phonetic cues to verbal meaning. To evaluate the success of this method, we studied the perception of corresponding speech audio and transformed EGG signals in an affect rating experiment with online listeners. The results show a high degree of similarity in the perceived affect of matched signals, indicating that our method is effective.

Classification of voice quality using neck-surface acceleration: Comparison with glottal flow and radiated sound

The aim of the present study is to investigate the usefulness of features extracted from miniature accelerometers attached to speaker's tracheal wall below the glottis for classification of phonation type. The performance of the accelerometer features is evaluated relative to features obtained from inverse filtered and radiated sound. While the former is a good proxy for the voice source, obtaining robust voice source features from the latter is considered difficult since it also contains information about the vocal tract filter. By contrast, the accelerometer signal is largely unaffected by the vocal tract and although it is shaped by subglottal resonances and the transfer properties of the neck tissue, these properties remain constant within a speaker. For this reason, we expect it to provide a better approximation of the voice source than the raw audio. We also investigate which aspects of the voice source are derivable from the accelerometer and microphone signals. Five trained singers (two females and three males) were recorded producing the syllable [pæ:] in three voice qualities (neutral, breathy and pressed) and at three pitch levels as determined by the participants' personal preference. Features extracted from the three signals were used for classification of phonation type using a random forest classifier. In addition, accelerometer and microphone features with highest correlation with the voice source features were identified. The three signals showed comparable classification error rates, with considerable differences across speakers both with respect to the overall performance and the importance of individual features. The speaker-specific differences notwithstanding, variation of phonation type had consistent effects on the voice source, accelerometer and audio signals. With regard to the voice source, AQ, NAQ, L1L2 and CQ all showed a monotonic variation along the breathy - neutral - pressed continuum. Several features were also found to vary systematically in the accelerometer and audio signals: HRF, L1L2 and CPPS (both the accelerometer and the audio), as well as the sound level (for the audio). The random forest analysis revealed that all of these features were also among the most important for the classification of voice quality. Both the accelerometer and the audio signals were found to discriminate between phonation types with an accuracy approaching that of the voice source. Thus, the accelerometer signal, which is largely uncontaminated by vocal tract resonances, offered no advantage over the signal collected with a normal microphone.

Effects of Spectral Shaping on Speech Auditory Brainstem Responses to Stop Consonant-Vowel Syllables.

Spectral shaping is employed by hearing aids to make consonantal information, such as formant transitions, audible for listeners with hearing loss. How manipulations of the stimuli, such as spectral shaping, may alter encoding in the auditory brainstem has not been thoroughly studied. The aim of this study was to determine how spectral shaping of synthetic consonant-vowel (CV) syllables, varying in their second formant (F2) onset frequency, may affect encoding of the syllables in the auditory brainstem. We employed a repeated measure design. Sixteen young adults (mean = 20.94 years, 6 males) and 11 older adults (mean = 58.60 years, 4 males) participated in this study. Speech-evoked auditory brainstem responses (speech-ABRs) were obtained from each participant using three CV exemplars selected from synthetic stimuli generated for a /ba-da-ga/ continuum. Brainstem responses were also recorded to corresponding three CV exemplars that were spectrally shaped to decrease low-frequency information and provide gain for middle and high frequencies according to a Desired Sensation Level function. In total, six grand average waveforms (3 phonemes [/ba/, /da/, /ga/] X 2 shaping conditions [unshaped, shaped]) were produced for each participant. Peak latencies and amplitudes, referenced to prestimulus baseline, were identified for 15 speech-ABR peaks. Peaks were marked manually using the program cursor on each individual waveform. Repeated-measures analysis of variances were used to determine the effects of shaping on the latencies and amplitudes of the speech-ABR peaks. Shaping effects produced changes within participants in ABR latencies and amplitudes involving onset and major peaks of the speech-ABR waveform for certain phonemes. Specifically, data from onset peaks showed that shaping decreased latency for /ga/ in older listeners, and decreased amplitude onset for /ba/ in younger listeners. Shaping also increased the amplitudes of major peaks for /ga/ stimuli in both groups. Encoding of speech in the ABR waveform may be more complex and multidimensional than a simple demarcation of source and filter information, and may also be influenced by cue intensity and age. These results suggest a more complex subcortical encoding of vocal tract filter information in the ABR waveform, which may also be influenced by cue intensity and age.

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.

Analysis and classification of speech sounds of children with autism spectrum disorder using acoustic features

Children with autism spectrum disorder (ASD) produce speech sounds different from that of Normal or non-ASD children. Hence, analyzing acoustic features can help characterizing the ASD speech signals. In this study, the distinguishing characteristics of speech production are examined for ASD affected children, with comparison to Normal children’s speech. Acoustic features are analyzed first and then classification of ASD vs Normal speech is attempted using different machine learning techniques. Two speech sound databases are recorded for this study: the speech database of children with ASD and the speech database of Normal children. English speech utterances are recorded for children of Indian regional (Tamil and Telugu) nativity. The changes due to autism effect are examined in context of 5 English vowel sounds (/a/, /e/, /i/, /o/, and /u/). Changes in the speech production characteristics of children are explored using three sets of features. Firstly, changes in the excitation source features are examined using strength of excitation (SoE) and instantaneous fundamental frequency (F0). Secondly, changes in the vocal tract (VT) filter features are examined using dominant frequencies (FD1, FD2) and formant frequencies (F1 to F5). Thirdly, changes in the source-filter combined features are examined using signal energy (E), zero-crossing rate (ZCR), linear prediction cepstrum coefficients (LPCC), and Mel-frequency cepstral coefficients (MFCC). Then, various combinations of the acoustic features are classified utilizing machine learning methods such as probabilistic neural network (PNN), multilayer perceptron (MLP), support vector machine (SVM), and K-nearest neighbors (KNN). Analyses of acoustic features shows significant differences between the speech of children with ASD and the Normal children. Results up to 98.17% accuracy are obtained for classification between acoustic features of the speech sounds of children with ASD and the Normal children. The observations and this study results may be useful as acoustic biomarkers to identify autism and its progression/cure among children. This study may also be valuable towards developing a system for ASD diagnosis from children’s speech sounds, in the future.

Using browser-based software to teach acoustic theory of speech production

I will demonstrate how I use browser-based software authored by Mark Huckvale to promote active learning of the acoustic theory of speech production. Students create the complex glottal tone using ESynth, filter the glottal tone with various vocal tract filters with ESystem, explore the impact of vocal tract shapes on formants using VTDemo, and examine their own vowel formants with ESection.

The evolution of the pectoral extrinsic appendicular and infrahyoid musculature in theropods and its functional and behavioral importance.

Birds have lost and modified the musculature joining the pectoral girdle to the skull and hyoid, called the pectoral extrinsic appendicular and infrahyoid musculature. These muscles include the levator scapulae, sternomandibularis, sternohyoideus, episternocleidomastoideus, trapezius, and omohyoideus. As non-avian theropod dinosaurs are the closest relatives to birds, it is worth investigating what conditions they may have exhibited to learn when and how these muscles were lost or modified. Using extant phylogenetic bracketing, osteological correlates and non-osteological influences of these muscles are identified and discussed. Compsognathids and basal Maniraptoriformes were found to have been the likeliest transition points of a derived avian condition of losing or modifying these muscles. Increasing needs to control the feather tracts of the neck and shoulder, for insulation, display, or tightening/readjustment of the skin after dynamic neck movements may have been the selective force that drove some of these muscles to be modified into dermo-osseous muscles. The loss and modification of shoulder protractors created a more immobile girdle that would later be advantageous for flight in birds. The loss of the infrahyoid muscles freed the hyolarynx, trachea, and esophagus which may have aided in vocal tract filtering.

Effect of the vocal tract filtering on the measurement of vocal cycle perturbations

Jitter and shimmer appear as symptoms of disordered voices and they are defined as the cycle-to-cycle variation of the cycle length and cycle amplitude, respectively. The frequency modulation of the phonatory excitation, i.e. vocal jitter, causes amplitude modulation (shimmer) of the speech signal due to the action of the vocal tract transfer function. In this Letter, the authors investigate the extent to which the vocal tract filtering affects the accuracy of measurement of vocal cycle perturbations of the phonatory excitation. To achieve this goal, the direct digital synthesiser is used to generate a series of glottal excitation cycles with the desired amount of vocal perturbations that crosses the vocal tract model to produce a set of 231 stimuli of sustained vowels [a]. Praat is used to measure the amount of perturbations. Results show that the values of the input shimmer inserted in the glottal excitation are smaller than the values measured by Praat. An undesired amount of shimmer is generated in the speech signal when the glottal excitation signal perturbed with jitter crosses through the vocal tract.

Epoch Detection Using Hilbert Envelope for Glottal Excitation Enhancement and Maximum-Sum Subarray for Epoch Marking

A technique is presented for detecting glottal excitation epochs in the speech signal, using Hilbert envelope for enhancing the glottal excitation and maximum-sum subarray for marking the epochs. The processing comprises the steps of dynamic range compression, Hilbert envelope calculation, saliency enhancing, and epoch marking. The dynamic range compression reduces the amplitude variation of the signal. The Hilbert envelope enhances the glottal excitation. The saliency enhancing further enhances the instants of significant excitation by reducing the residual ripples related to the vocal tract filter, by using a dynamic peak detector, a nonlinear smoother, and a differentiator. The epoch marking locates the peak of the maximum-sum subarray as the instant of significant glottal excitation. Evaluation of the proposed technique showed its performance measures to be similar to those of the state-of-the-art techniques for normal speech, telephone-quality speech, and highpass filtered speech and better than or similar to them for pathological speech. The averaged accuracy-weighted identification rates with the proposed technique for normal speech, telephone-quality speech, and pathological speech were 79.46%, 77.04%, and 71.52%, respectively. The proposed technique employs single-pass processing and may find applications in speech training aids, diagnosis of speech disorders, and voice conversion of speech with voice disorders.

Experimental study on nonlinear source-filter interaction using synthetic vocal fold models.

Under certain conditions, e.g., singing voice, the fundamental frequency of the vocal folds can go up and interfere with the formant frequencies. Acoustic feedback from the vocal tract filter to the vocal fold source then becomes strong and non-negligible. An experimental study was presented on such source-filter interaction using three types of synthetic vocal fold models. Asymmetry was also created between the left and right vocal folds. The experiment reproduced various nonlinear phenomena, such as frequency jump and quenching, as reported in humans. Increase in phonation threshold pressure was also observed when resonant frequency of the vocal tract and fundamental frequency of the vocal folds crossed each other. As a combined effect, the phonation threshold pressure was further increased by the left-right asymmetry. Simulation of the asymmetric two-mass model reproduced the experiments to some extent. One of the intriguing findings of this study is the variable strength of the source-filter interaction over different model types. Among the three models, two models were strongly influenced by the vocal tract, while no clear effect of the vocal tract was observed in the other model. This implies that the level of source-filter interaction may vary considerably from one subject to another in humans.

OPENGLOT – An open environment for the evaluation of glottal inverse filtering

Glottal inverse filtering (GIF) refers to technology to estimate the source of voiced speech, the glottal flow, from speech signals. When a new GIF algorithm is proposed, its accuracy needs to be evaluated. However, the evaluation of GIF is problematic because the ground truth, the real glottal volume velocity signal generated by the vocal folds, cannot be recorded non-invasively from natural speech. This absence of the ground truth has been circumvented in most previous GIF studies by using simple linear source-filter synthesis techniques with known artificial glottal flow models and all-pole vocal tract filters. Moreover, in a few previous studies, physical modeling of speech production has been utilized in synthesis of the test data for GIF evaluation. The evaluation strategy in previous GIF studies is, however, scattered between individual investigations and there is currently a lack of a coherent, common platform to be used in GIF evaluation. In order to address this shortcoming, the current study introduces a new environment, called OPENGLOT, for GIF evaluation. The key ideas of OPENGLOT are twofold: the environment is versatile (i.e., it provides different types of test signals for GIF evaluation) and open (i.e., the system can be used by anyone who wants to evaluate her or his new GIF method and compare it objectively to previously developed benchmark techniques). OPENGLOT consists of four main parts, Repositories I–IV, that contain data and sound synthesis software. Repository I contains a large set of synthetic glottal flow waveforms, and speech signals generated by using the Liljencrants–Fant (LF) waveform as an artificial excitation, and a digital all-pole filter to model the vocal tract. Repository II contains glottal flow and speech pressure signals generated using physical modeling of human speech production. Repository III contains pairs of glottal excitation and speech pressure signal generated by exciting 3D printed plastic vocal tract replica with LF excitations via a loudspeaker. Finally, Repository IV contains multichannel recordings (speech pressure signal, electroglottogram, high-speed video of the vocal folds) from natural production of speech. After presenting these four core parts of OPENGLOT, the article demonstrates the platform by presenting a typical use case.

Speech spoofing countermeasures based on source voice analysis and machine learning techniques

Automatic speaker verification (ASV) [7] systems are susceptible to malicious attacks. It discredit the performance of a standard ASV system by increasing its false acceptance rates. This paper presents a new countermeasure for the protection of automatic speaker verification systems from spoofed signals. The new countermeasure is based on the analysis of a sequence of acoustic feature vectors using the glottal inverse filtering. In the proposed method, speech is decomposed into a glottal source signal and model the vocal tract filter through glottal inverse filtering. The IAIF desriptors are constructed and are used as features. Support Vector Machines (SVM) classifier and Extreme learning machine (ELM) are used to classify the obtained features as genuine or spoofed. It is hoped that the proposed method can help to detect the genuine speech from the spoofed one.

Accurate Computation of Vocal Tract Filter Parameters Using a Hybrid Genetic Algorithm

ABSTRACTThis paper reports an original technique for accurately estimating the parameters of human vocal tract filters for vowels in English for speech processing applications such as voice recognition. In this paper, the vocal tract filter design problem is reformulated as a general nonlinear optimization problem and solved using a hybrid Genetic Algorithm (GA). The hybrid GA computes a rough estimate of the global minimum using GA and refines using computationally cheap local search. Issues that are of concern in digital filtering such as achieving stability and overcoming finite precision effects are addressed. The objective function for optimization used in this paper is formulated in terms of poles and zeros of the filters to avoid ill-conditioning and to take advantage of symmetries in the location of poles and zeros. Simulation results indicate that the approach presented in this paper provides a close fit in terms of mean square error between the experimental and designed filters.

Limitations of the source-filter coupling in phonation

The coupling of vocal fold (source) and vocal tract (filter) is one of the most critical factors in source-filter articulation theory. The traditional linear source-filter theory has been challenged by current research which clearly shows the impact of acoustic loading on the dynamic behavior of the vocal fold vibration as well as the variations in the glottal flow pulses shape. This paper outlines the underlying mechanism of source-filter interactions; demonstrates the design and working principles of coupling for the various existing vocal cord and vocal tract biomechanical models. For our study, we have considered self-oscillating lumped-element models of the acoustic source and computational models of the vocal tract as articulators. To understand the limitations of source-filter interactions which are associated with each of those models, we compare them concerning their mechanical design, acoustic and physiological characteristics and aerodynamic simulation.

Speaker recognition from whispered speech: A tutorial survey and an application of time-varying linear prediction

From the available biometric technologies, automatic speaker recognition is one of the most convenient and accessible ones due to abundance of mobile devices equipped with a microphone, allowing users to be authenticated across multiple environments and devices. Speaker recognition also finds use in forensics and surveillance. Due to the acoustic mismatch induced by varied environments and devices of the same speaker, leading to increased number of identification errors, much of the research focuses on compensating for such technology-induced variations, especially using machine learning at the statistical back-end. Another much less studied but at least as detrimental source of acoustic variation, however, arises from mismatched speaking styles induced by the speaker, leading to a substantial performance drop in recognition accuracy. This is a major problem especially in forensics where perpetrators may purposefully disguise their identity by varying their speaking style. We focus on one of the most commonly used ways of disguising one’s speaker identity, namely, whispering. We approach the problem of normal-whisper acoustic mismatch compensation from the viewpoint of robust feature extraction. Since whispered speech is intelligible, yet a low-intensity signal and therefore prone to extrinsic distortions, we take advantage of robust, long-term speech analysis methods that utilize slow articulatory movements in speech production. In specific, we address the problem using a novel method, frequency-domain linear prediction with time-varying linear prediction (FDLP-TVLP), which is an extension of the 2-dimensional autoregressive (2DAR) model that allows vocal tract filter parameters to be time-varying, rather than piecewise constant as in classic short-term speech analysis. Our speaker recognition experiments on the whisper subset of the CHAINS corpus indicate that when tested in normal-whisper mismatched conditions, the proposed FDLP-TVLP features improve speaker recognition performance by 7–10% over standard MFCC features in relative terms. We further observe that the proposed FDLP-TVLP features perform better than the FDLP and 2DAR methods for whispered speech.