Multi-parametric source-filter separation of speech and prosodic voice restoration

In this thesis, methods and models are developed and presented aiming at the estimation, restoration and transformation of the characteristics of human speech. During a first period of the thesis, a concept was developed that allows restoring prosodic voice features and reconstruct more natural sounding speech from pathological voices using a multi-resolution approach. Inspired from observations with respect to this approach, the necessity of a novel method for the separation of speech into voice source and articulation components emerged in order to improve the perceptive quality of the restored speech signal. This work subsequently represents the main part of this work and therefore is presented first in this thesis. The proposed method is evaluated on synthetic, physically modelled, healthy and pathological speech. A robust, separate representation of source and filter characteristics has applications in areas that go far beyond the reconstruction of alaryngeal speech. It is potentially useful for efficient speech coding, voice biometrics, emotional speech synthesis, remote and/or non-invasive voice disorder diagnosis, etc. A key aspect of the voice restoration method is the reliable separation of the speech signal into voice source and articulation for it is mostly the voice source that requires replacement or enhancement in alaryngeal speech. Observations during the evaluation of above method highlighted that this separation is insufficient with currently known methods. Therefore, the main part of this thesis is concerned with the modelling of voice and vocal tract and the estimation of the respective model parameters. Most methods for joint source filter estimation known today represent a compromise between model complexity, estimation feasibility and estimation efficiency. Typically, single-parametric models are used to represent the source for the sake of tractable optimization or multi-parametric models are estimated using inefficient grid searches over the entire parameter space. The novel method presented in this work proposes advances in the direction of efficiently estimating and fitting multi-parametric source and filter models to healthy and pathological speech signals, resulting in a more reliable estimation of voice source and especially vocal tract coefficients. In particular, the proposed method is exhibits a largely reduced bias in the estimated formant frequencies and bandwidths over a large variety of experimental conditions such as environmental noise, glottal jitter, fundamental frequency, voice types and glottal noise. The methods appears to be especially robust to environmental noise and improves the separation of deterministic voice source components from the articulation. Alaryngeal speakers often have great difficulty at producing intelligible, not to mention prosodic, speech. Despite great efforts and advances in surgical and rehabilitative techniques, currently known methods, devices and modes of speech rehabilitation leave pathological speakers with a lack in the ability to control key aspects of their voice. The proposed multiresolution approach presented at the end of this thesis provides alaryngeal speakers an intuitive manner to increase prosodic features in their speech by reconstructing a more intelligible, more natural and more prosodic voice. The proposed method is entirely non-invasive. Key prosodic cues are reconstructed and enhanced at different temporal scales by inducing additional volatility estimated from other, still intact, speech features. The restored voice source is thus controllable in an intuitive way by the alaryngeal speaker. Despite the above mentioned advantages there is also a weak point of the proposed joint source-filter estimation method to be mentioned. The proposed method exhibits a susceptibility to modelling errors of the glottal source. On the other hand, the proposed estimation framework appears to be well suited for future research on exactly this topic. A logical continuation of this work is the leverage the efficiency and reliability of the proposed method for the development of new, more accurate glottal source models.

HMM mixtures (HMM2) for robust speech recognition

Katrin Weber

State-of-the-art automatic speech recognition (ASR) techniques are typically based on hidden Markov models (HMMs) for the modeling of temporal sequences of feature vectors extracted from the speech signal. At the level of each HMM state, Gaussian mixture models (GMMs) or artificial neural networks (ANNs) are commonly used in order to model the state emission probabilities. However, both GMMs and ANNs are rather rigid, as they are incapable of adapting to variations inherent in the speech signal, such as inter- and intra-speaker variations. Moreover, performance degradations of these systems are severe in the case of unmatched conditions such as in the presence of environmental noise. A lot of research effort is currently being devoted to overcoming these problems. The principal objective of this thesis is to explore new approaches towards a more robust and adaptive modeling of speech. In this context, different aspects of the modeling of speech data with HMMs and GMMs are investigated. Particular attention is given to the modeling of correlation. While correlation between different feature vectors (corresponding to temporal correlation) is typically modeled by the HMM, correlation between feature vector components (e.g., correlation in frequency) is modeled by the GMM part of the model. This thesis starts with the investigation of two potential ways to improve the modeling of correlation, consisting of (1) a shift of the modeling of temporal correlation towards GMMs, and (2) the modeling of correlation within each feature vector by a particular type of HMM. This leads to the development of a novel approach, referred to as "HMM2", which is a major focus of this thesis. HMM2 is a particular mixture of hidden Markov models, where state emission probabilities of the temporal (primary) HMM are modeled through (secondary) state-dependent frequency-based HMMs. Low-dimensional GMMs are used for modeling the state emission probabilities of the secondary HMM states. Therefore, HMM2 can be seen as a generalization of conventional HMMs, which they include as a particular case. HMM2 may have several advantages as compared to standard systems. While the primary HMM performs time warping and time integration, the secondary HMM performs warping and integration along the frequency dimension of the speech signal. Frequency correlation is modeled through the secondary HMM topology. Due to the implicit, non-linear, state-dependent spectral warping performed by the secondary HMM, HMM2 may be viewed as a dynamic extension of the multi-band approach. Moreover, this frequency warping property may result in a better, more flexible modeling and parameter sharing. After an investigation of theoretical and practical aspects of HMM2, encouraging recognition results for the case of speech degraded by additive noise are given. Due to the spectral warping property of HMM2, this model is able to extract pertinent structural information of the speech signal, which is reflected in the trained model parameters. Consequently, such an HMM2 system can also be used to explicitly extract structures of a speech signal, which can then be converted into a new kind of ASR features, referred to as "HMM2 features". In fact, frequency bands with similar characteristics are supposed to be emitted by the same secondary HMM state. The warping along the frequency dimension of speech thus results in an adaptable, data-driven frequency segmentation. In fact, as it can be assumed that different secondary HMM states model spectral regions characterized by high and low energies respectively, this segmentation may be related to formant structures. The application of HMM2 as a feature extractor is investigated, and it is shown that a system combining HMM2 features with conventional noise-robust features yields an improved speech recognition robustness. Moreover, a comparison of HMM2 features with formant tracks shows a comparable performance on a vowel classification task.

EPFL2003

HMM Mixtures (HMM2) for Robust Speech Recognition

Katrin Weber

State-of-the-art automatic speech recognition (ASR) techniques are typically based on hidden Markov models (HMMs) for the modeling of temporal sequences of feature vectors extracted from the speech signal. At the level of each HMM state, Gaussian mixture models (GMMs) or artificial neural networks (ANNs) are commonly used in order to model the state emission probabilities. However, both GMMs and ANNs are rather rigid, as they are incapable of adapting to variations inherent in the speech signal, such as inter- and intra-speaker variations. Moreover, performance degradations of these systems are severe in the case of unmatched conditions such as in the presence of environmental noise. A lot of research effort is currently being devoted to overcoming these problems. The principal objective of this thesis is to explore new approaches towards a more robust and adaptive modeling of speech. In this context, different aspects of the modeling of speech data with HMMs and GMMs are investigated. Particular attention is given to the modeling of correlation. While correlation between different feature vectors (corresponding to temporal correlation) is typically modeled by the HMM, correlation between feature vector components (e.g., correlation in frequency) is modeled by the GMM part of the model. This thesis starts with the investigation of two potential ways to improve the modeling of correlation, consisting of (1) a shift of the modeling of temporal correlation towards GMMs, and (2) the modeling of correlation within each feature vector by a particular type of HMM. This leads to the development of a novel approach, referred to as ÒHMM2Ó, which is a major focus of this thesis. HMM2 is a particular mixture of hidden Markov models, where state emission probabilities of the temporal (primary) HMM are modeled through (secondary) state-dependent frequency-based HMMs. Low-dimensional GMMs are used for modeling the state emission probabilities of the secondary HMM states. Therefore, HMM2 can be seen as a generalization of conventional HMMs, which they include as a particular case. HMM2 may have several advantages as compared to standard systems. While the primary HMM performs time warping and time integration, the secondary HMM performs warping and integration along the frequency dimension of the speech signal. Frequency correlation is modeled through the secondary HMM topology. Due to the implicit, non-linear, state-dependent spectral warping performed by the secondary HMM, HMM2 may be viewed as a dynamic extension of the multi-band approach. Moreover, this frequency warping property may result in a better, more flexible modeling and parameter sharing. After an investigation of theoretical and practical aspects of HMM2, encouraging recognition results for the case of speech degraded by additive noise are given. Due to the spectral warping property of HMM2, this model is able to extract pertinent structural information of the speech signal, which is reflected in the trained model parameters. Consequently, such an HMM2 system can also be used to explicitly extract structures of a speech signal, which can then be converted into a new kind of ASR features, referred to as ÒHMM2 featuresÓ. In fact, frequency bands with similar characteristics are supposed to be emitted by the same secondary HMM state. The warping along the frequency dimension of speech thus results in an adaptable, data-driven frequency segmentation. In fact, as it can be assumed that different secondary HMM states model spectral regions characterized by high and low energies respectively, this segmentation may be related to formant structures. The application of HMM2 as a feature extractor is investigated, and it is shown that a system combining HMM2 features with conventional noise-robust features yields an improved speech recognition robustness. Moreover, a comparison of HMM2 features with formant tracks shows a comparable performance on a vowel classification task.

Ecole Polytechnique Federale de Lausanne2003

HMM Mixtures (HMM2) for Robust Speech Recognition

Katrin Weber

State-of-the-art automatic speech recognition (ASR) techniques are typically based on hidden Markov models (HMMs) for the modeling of temporal sequences of feature vectors extracted from the speech signal. At the level of each HMM state, Gaussian mixture models (GMMs) or artificial neural networks (ANNs) are commonly used in order to model the state emission probabilities. However, both GMMs and ANNs are rather rigid, as they are incapable of adapting to variations inherent in the speech signal, such as inter- and intra-speaker variations. Moreover, performance degradations of these systems are severe in the case of unmatched conditions such as in the presence of environmental noise. A lot of research effort is currently being devoted to overcoming these problems. The principal objective of this thesis is to explore new approaches towards a more robust and adaptive modeling of speech. In this context, different aspects of the modeling of speech data with HMMs and GMMs are investigated. Particular attention is given to the modeling of correlation. While correlation between different feature vectors (corresponding to temporal correlation) is typically modeled by the HMM, correlation between feature vector components (e.g., correlation in frequency) is modeled by the GMM part of the model. This thesis starts with the investigation of two potential ways to improve the modeling of correlation, consisting of (1) a shift of the modeling of temporal correlation towards GMMs, and (2) the modeling of correlation within each feature vector by a particular type of HMM. This leads to the development of a novel approach, referred to as ÒHMM2Ó, which is a major focus of this thesis. HMM2 is a particular mixture of hidden Markov models, where state emission probabilities of the temporal (primary) HMM are modeled through (secondary) state-dependent frequency-based HMMs. Low-dimensional GMMs are used for modeling the state emission probabilities of the secondary HMM states. Therefore, HMM2 can be seen as a generalization of conventional HMMs, which they include as a particular case. HMM2 may have several advantages as compared to standard systems. While the primary HMM performs time warping and time integration, the secondary HMM performs warping and integration along the frequency dimension of the speech signal. Frequency correlation is modeled through the secondary HMM topology. Due to the implicit, non-linear, state-dependent spectral warping performed by the secondary HMM, HMM2 may be viewed as a dynamic extension of the multi-band approach. Moreover, this frequency warping property may result in a better, more flexible modeling and parameter sharing. After an investigation of theoretical and practical aspects of HMM2, encouraging recognition results for the case of speech degraded by additive noise are given. Due to the spectral warping property of HMM2, this model is able to extract pertinent structural information of the speech signal, which is reflected in the trained model parameters. Consequently, such an HMM2 system can also be used to explicitly extract structures of a speech signal, which can then be converted into a new kind of ASR features, referred to as ÒHMM2 featuresÓ. In fact, frequency bands with similar characteristics are supposed to be emitted by the same secondary HMM state. The warping along the frequency dimension of speech thus results in an adaptable, data-driven frequency segmentation. In fact, as it can be assumed that different secondary HMM states model spectral regions characterized by high and low energies respectively, this segmentation may be related to formant structures. The application of HMM2 as a feature extractor is investigated, and it is shown that a system combining HMM2 features with conventional noise-robust features yields an improved speech recognition robustness. Moreover, a comparison of HMM2 features with formant tracks shows a comparable performance on a vowel classification task.

IDIAP2003

HMM mixtures (HMM2) for robust speech recognition

Katrin Weber

EPFL2003

HMM Mixtures (HMM2) for Robust Speech Recognition

Katrin Weber

State-of-the-art automatic speech recognition (ASR) techniques are typically based on hidden Markov models (HMMs) for the modeling of temporal sequences of feature vectors extracted from the speech signal. At the level of each HMM state, Gaussian mixture models (GMMs) or artificial neural networks (ANNs) are commonly used in order to model the state emission probabilities. However, both GMMs and ANNs are rather rigid, as they are incapable of adapting to variations inherent in the speech signal, such as inter- and intra-speaker variations. Moreover, performance degradations of these systems are severe in the case of unmatched conditions such as in the presence of environmental noise. A lot of research effort is currently being devoted to overcoming these problems. The principal objective of this thesis is to explore new approaches towards a more robust and adaptive modeling of speech. In this context, different aspects of the modeling of speech data with HMMs and GMMs are investigated. Particular attention is given to the modeling of correlation. While correlation between different feature vectors (corresponding to temporal correlation) is typically modeled by the HMM, correlation between feature vector components (e.g., correlation in frequency) is modeled by the GMM part of the model. This thesis starts with the investigation of two potential ways to improve the modeling of correlation, consisting of (1) a shift of the modeling of temporal correlation towards GMMs, and (2) the modeling of correlation within each feature vector by a particular type of HMM. This leads to the development of a novel approach, referred to as ÒHMM2Ó, which is a major focus of this thesis. HMM2 is a particular mixture of hidden Markov models, where state emission probabilities of the temporal (primary) HMM are modeled through (secondary) state-dependent frequency-based HMMs. Low-dimensional GMMs are used for modeling the state emission probabilities of the secondary HMM states. Therefore, HMM2 can be seen as a generalization of conventional HMMs, which they include as a particular case. HMM2 may have several advantages as compared to standard systems. While the primary HMM performs time warping and time integration, the secondary HMM performs warping and integration along the frequency dimension of the speech signal. Frequency correlation is modeled through the secondary HMM topology. Due to the implicit, non-linear, state-dependent spectral warping performed by the secondary HMM, HMM2 may be viewed as a dynamic extension of the multi-band approach. Moreover, this frequency warping property may result in a better, more flexible modeling and parameter sharing. After an investigation of theoretical and practical aspects of HMM2, encouraging recognition results for the case of speech degraded by additive noise are given. Due to the spectral warping property of HMM2, this model is able to extract pertinent structural information of the speech signal, which is reflected in the trained model parameters. Consequently, such an HMM2 system can also be used to explicitly extract structures of a speech signal, which can then be converted into a new kind of ASR features, referred to as ÒHMM2 featuresÓ. In fact, frequency bands with similar characteristics are supposed to be emitted by the same secondary HMM state. The warping along the frequency dimension of speech thus results in an adaptable, data-driven frequency segmentation. In fact, as it can be assumed that different secondary HMM states model spectral regions characterized by high and low energies respectively, this segmentation may be related to formant structures. The application of HMM2 as a feature extractor is investigated, and it is shown that a system combining HMM2 features with conventional noise-robust features yields an improved speech recognition robustness. Moreover, a comparison of HMM2 features with formant tracks shows a comparable performance on a vowel classification task.

Ecole Polytechnique Federale de Lausanne2003

HMM Mixtures (HMM2) for Robust Speech Recognition

Katrin Weber

State-of-the-art automatic speech recognition (ASR) techniques are typically based on hidden Markov models (HMMs) for the modeling of temporal sequences of feature vectors extracted from the speech signal. At the level of each HMM state, Gaussian mixture models (GMMs) or artificial neural networks (ANNs) are commonly used in order to model the state emission probabilities. However, both GMMs and ANNs are rather rigid, as they are incapable of adapting to variations inherent in the speech signal, such as inter- and intra-speaker variations. Moreover, performance degradations of these systems are severe in the case of unmatched conditions such as in the presence of environmental noise. A lot of research effort is currently being devoted to overcoming these problems. The principal objective of this thesis is to explore new approaches towards a more robust and adaptive modeling of speech. In this context, different aspects of the modeling of speech data with HMMs and GMMs are investigated. Particular attention is given to the modeling of correlation. While correlation between different feature vectors (corresponding to temporal correlation) is typically modeled by the HMM, correlation between feature vector components (e.g., correlation in frequency) is modeled by the GMM part of the model. This thesis starts with the investigation of two potential ways to improve the modeling of correlation, consisting of (1) a shift of the modeling of temporal correlation towards GMMs, and (2) the modeling of correlation within each feature vector by a particular type of HMM. This leads to the development of a novel approach, referred to as ÒHMM2Ó, which is a major focus of this thesis. HMM2 is a particular mixture of hidden Markov models, where state emission probabilities of the temporal (primary) HMM are modeled through (secondary) state-dependent frequency-based HMMs. Low-dimensional GMMs are used for modeling the state emission probabilities of the secondary HMM states. Therefore, HMM2 can be seen as a generalization of conventional HMMs, which they include as a particular case. HMM2 may have several advantages as compared to standard systems. While the primary HMM performs time warping and time integration, the secondary HMM performs warping and integration along the frequency dimension of the speech signal. Frequency correlation is modeled through the secondary HMM topology. Due to the implicit, non-linear, state-dependent spectral warping performed by the secondary HMM, HMM2 may be viewed as a dynamic extension of the multi-band approach. Moreover, this frequency warping property may result in a better, more flexible modeling and parameter sharing. After an investigation of theoretical and practical aspects of HMM2, encouraging recognition results for the case of speech degraded by additive noise are given. Due to the spectral warping property of HMM2, this model is able to extract pertinent structural information of the speech signal, which is reflected in the trained model parameters. Consequently, such an HMM2 system can also be used to explicitly extract structures of a speech signal, which can then be converted into a new kind of ASR features, referred to as ÒHMM2 featuresÓ. In fact, frequency bands with similar characteristics are supposed to be emitted by the same secondary HMM state. The warping along the frequency dimension of speech thus results in an adaptable, data-driven frequency segmentation. In fact, as it can be assumed that different secondary HMM states model spectral regions characterized by high and low energies respectively, this segmentation may be related to formant structures. The application of HMM2 as a feature extractor is investigated, and it is shown that a system combining HMM2 features with conventional noise-robust features yields an improved speech recognition robustness. Moreover, a comparison of HMM2 features with formant tracks shows a comparable performance on a vowel classification task.

IDIAP2003