Graphical models provide a promising paradigm to study both existing and novel techniques for automatic speech recognition. This paper first provides a brief overview of graphical models and their uses as statistical models. It is then shown that the statistical assumptions behind many pattern recognition techniques commonly used as part of a speech recognition system can be described by a graph – this includes Gaussian distributions, mixture models, decision trees, factor analysis, principle component analysis, linear discriminant analysis, and hidden Markov models. Moreover, this paper shows that many advanced models for speech recognition and language processing can also be simply described by a graph, including many at the acoustic-, pronunciation-, and language-modeling levels. A number of speech recognition techniques born directly out of the graphical-models paradigm are also surveyed. Additionally, this paper includes a novel graphical analysis regarding why derivative (or delta) features improve hidden Markov model-based speech recognition by improving structural discriminability. It also includes an example where a graph can be used to represent language model smoothing constraints. As will be seen, the space of models describable by a graph is quite large. A thorough exploration of this space should yield techniques that ultimately will supersede the hidden Markov model.

Current-generation automatic speech recognition #ASR# systems assume that words are readily decomposable into constituent phonetic components ##phonemes"#. A detailed linguistic dissection of state-of-the-art speech recognition systems indicates that the conventional phonemic #beads-on-a-string" approach is of limited utility, particularly with respect to informal, conversational material. The study shows that there is a signi#cantgapbetween the observed data and the pronunciation models of current ASR systems. It also shows that many important factors a#ecting recognition performance are not modeled explicitly in these systems.

Declaration This dissertation is the result of my own work and includes nothing which is the outcome of work done in collaboration, except where stated. It has not been submitted in whole or part for a degree at any other university. The length of this thesis including footnotes and appendices does not exceed 29,500 words and includes no more than 40 figures. 1 Systems which automatically transcribe carefully dictated speech are now commercially available, but their performance degrades dramatically when the speaking style of users becomes more relaxed or conversational. This dissertation focuses on techniques that aim to improve the robustness of statistical speech transcription systems to conversational speaking styles. The dissertation shows first that the performance degradation occuring as speech becomes more conversational is severe and is partially attributable to differences in the acoustic realizations of sentences. Hypothesizing that the quantifiably wider range of

Microphones and cameras have been extensively used to observe and detect human activity and to facilitate natural modes of interaction between humans and intelligent systems. Human brain processes the audio and video modalities extracting complementary and robust information from them. Intelligent systems with audio-visual sensors should be capable of achieving similar goals. The audio-visual information fusion strategy is a key component in designing such systems. In this paper we exclusively survey the fusion techniques used in various audio-visual information fusion tasks. The fusion strategy used tends to depend mainly on the model, probabilistic or otherwise, used in the particular task to process sensory information to obtain higher level semantic information. The models themselves are task oriented. In this paper we describe the fusion strategies and the corresponding models used in audiovisual tasks such as speech recognition, tracking, biometrics, affective state recognition and meeting scene analysis. We also review the challenges and existing solutions and also unresolved or partially resolved issues in these fields. Specifically, we discuss established and upcoming work in hierarchical fusion strategies and crossmodal learning techniques, identifying these as critical areas of research in the future development of intelligent systems.

In standard automatic speech recognition (ASR), hidden Markov models (HMMs) calculate their emission probabilities by an artificial neural network (ANN) or a Gaussian distribution conditioned only upon the hidden state variable. Recent work [12] showed the benefit of conditioning the emission distributions also upon a discrete auxiliary variable, which is observed in training and hidden in recognition. Related work [3] has shown the utility of conditioning the emission distributions on a continuous auxiliary variable. We apply mixed Bayesian networks (BNs) to extend these works by introducing a continuous auxiliary variable that is observed in training but is hidden in recognition. We find that an auxiliary pitch variable conditioned itself upon the hidden state can degrade performance unless the auxiliary variable is also hidden. The performance, furthermore, can be improved by making the auxiliary pitch variable independent of the hidden state.

Dissertation submitted to the University of Cambridge for the degree of Doctor of Philosophy Although the performance of speech recognition systems has increased substantially over the last decades, there still remain a number of tasks which pose considerable problems for current state-of-the-art techniques. One of these tasks is the recognition of spontaneous speech which differs from read or planned speech in that its underlying dynamics change frequently over time. The negative effect of changes in acoustic background condition on recognition performance can also be observed in other situations as, for instance, in the case of speech that is corrupted by non-stationary noise. This thesis is concerned with the development of an acoustic model for speech recognition which automatically detects changes in the background condition of a signal and compensates for the model-data mismatch by combining the information of several expert models. These experts are specialised on the different acoustic conditions under consideration and their influ-ence on the recognition process is determined by how well their associated condition matches

Automatic speech recognition (ASR) is a process of applying constraints, as encoded in the computer system (the recognizer), to the speech signal until ambiguity is satisfactorily resolved to the extent that only one sequence of words is hypothesized. Such constraints fall naturally into two categories. One deals with the the ordering of words (syntax) and organization of their meanings (semantics, pragmatics, etc). The other governs how speech signals are related to words, a process often termed as “lexical access”. This thesis studies the Huttenlocher-Zue lexical access model, its implementation in a modern probabilistic speech recognition framework and its application to continuous speech from an open vocabulary. The Huttenlocher-Zue model advocates a two-pass lexical access paradigm. In the first pass, the lexicon is effectively pruned using broad linguistic constraints. In the original Huttenlocher-Zue model, the authors had proposed six linguistic features motivated by the manner of pronunciation.

Confidence measures are computed to estimate the certainty that target acoustic units are spoken in specific speech segments. They are applied in tasks such as keyword verification or utterance verification. Because many of the confidence measures use the same set of models and features as in recognition, the resulting scores may not provide an independent measure of reliability. In this paper, we propose two articulatory feature (AF) based phoneme confidence measures that estimate the acoustic reliability based on the match in AF properties. While acoustic-based features, such as Mel-frequency cepstral coefficients (MFCC), are widely used in speech processing, some recent works have focus on linguistically based features, such as the articulatory features that relate directly to the human articulatory process which may better capture speech characteristics. The articulatory features can either replace or complement the acoustic-based features in speech processing. The proposed AF-based measures in this paper were evaluated, in comparison and in combination, with the HMM-based scores on phoneme and keyword verification tasks using childrenÕs speech collected for a computer-based English pronunciation learning project. To fully evaluate their usefulness, the proposed measures and combinations were evaluated on both native and non-native data; and under field test conditions that mis-matches with the training condition. The experimental results show that under the different environments, combinations of the AF scores with the HMM-based

. The challenge of automatic speech recognition (ASR) increases when speaker variability is encountered. Being able to automatically use dierent acoustic models according to speaker type might help to increase the robustness of ASR. We present a system that attempts to do so by augmenting the standard acoustic observations with pitch information. This allows the system to use acoustic models more appropriate to speech with the given pitch. Furthermore, pitch information is more easily detected in noisy conditions; thus, it may be of use in robust speech recognition. Using dynamic Bayesian networks (DBNs) allows further renement of the system by eliminating unnecessary statistical dependencies and thus reducing the number of parameters. We show that when a system is trained on observed pitch data and performs recognition with missing pitch data, it can perform signicantly better than a system that uses acoustics information only. Acknowledgements: Todd A. Stephenson and Mathew Magima...

Electropalatography (EPG) is a technique that determines the contact patterns between the tongue and the hard palate during speech. In one of its most common forms, it utilizes an artificial palate with 62 silver electrodes embedded in its tongue-facing surface. At small regular time intervals it is recorded whether a specific electrode is contacted or not by the tongue, leading to tongue-palate contact patterns.