This thesis addresses the significance of speech features within the task of speaker recognition. Motivated by the perception of simple attributes like `loud', `smooth', `fast', more than 70 new speech features are developed. A set of basic speech features like pitch, loudness and speech speed are combined together with these new features in a feature set, one set per utterance. A neural network classifier is used to evaluate the significance of these features by creating a speaker recognition system and analysing the behaviour of successfully trained single-speaker networks. An in-depth analysis of network weights allows a rating of significance and feature contribution. A subjective listening experiment validates and confirms the results of the neural network analysis. The work starts with an extended sentence analysis; ten sentences are uttered by 630 speakers. The extraction of 100 speech features is outlined and a 100-element feature vector for each utterance is derived. Some features themselves and the methods of analysing them have been used elsewhere, for example pitch, sound pressure level, spectral envelope, loudness, speech speed and glottal-to-noise excitation. However, more than 70 of the 100 features are derivatives of these basic features and have not yet been described and used before in the speakerr ecognition research,e speciallyyn ot within a rating of feature significance. These derivatives include histogram, 3`d and 4 moments, function approximation, as well as other statistical analyses applied to the basic features. The first approach assessing the significance of features and their possible use in a recognition system is based on a probability analysis. The analysis is established on the assumption that within the speaker's ten utterances' single feature values have a small deviation and cluster around the mean value of one speaker. The presented features indeed cluster into groups and show significant differences between speakers, thus enabling a clear separation of voices when applied to a small database of < 20 speakers. The recognition and assessment of individual feature contribution jecomes impossible, when the database is extended to 200 speakers. To ensure continous vplidation of feature contribution it is necessary to consider a different type of classifier. These limitations are overcome with the introduction of neural network classifiers. A separate network is assigned to each speaker, resulting in the creation of 630 networks. All networks are of standard feed-forward backpropagation type and have a 100-input, 20- hidden-nodes, one-output architecture. The 6300 available feature vectors are split into a training, validation and test set in the ratio of 5-3-2. The networks are initially trained with the same 100-feature input database. Successful training was achieved within 30 to 100 epochs per network. The speaker related to the network with the highest output is declared as the speaker represented by the input. The achieved recognition rate for 630 speakers is -49%. A subsequent preclusion of features with minor significance raises the recognition rate to 57%. The analysis of the network weight behaviour reveals two major pointsA definite ranking order of significance exists between the 100 features. Many of the newly introduced derivatives of pitch, brightness, spectral voice patterns and speech speed contribute intensely to recognition, whereas feature groups related to glottal-to-noiseexcitation ratio and sound pressure level play a less important role. The significance of features is rated by the training, testing and validation behaviour of the networks under data sets with reduced information content, the post-trained weight distribution and the standard deviation of weight distribution within networks. The findings match with results of a subjective listening experiment. As a second major result the analysis shows that there are large differences between speakers and the significance of features, i. e. not all speakers use the same feature set to the same extent. The speaker-related networks exhibit key features, where they are uniquely identifiable and these key features vary from speaker to speaker. Some features like pitch are used by all networks; other features like sound pressure level and glottal-to-noise excitation ratio are used by only a few distinct classifiers. Again, the findings correspond with results of a subjective listening experiment. This thesis presents more than 70 new features which never have been used before in speaker recognition. A quantitative ranking order of 100 speech features is introduced. Such a ranking order has not been documented elsewhere and is comparatively new to the area of speaker recognition. This ranking order is further extended and describes the amount to which a classifier uses or omits single features, solely depending on the characteristics of the voice sample. Such a separation has not yet been documented and is a novel contribution. The close correspondence of the subjective listening experiment and the findings of the network classifiers show that it is plausible to model the behaviour of human speech recognition with an artificial neural network. Again such a validation is original in the area of speaker recognition

published_or_final_version / Speech and Hearing Sciences / Master / Master of Philosophy

No description available.

No description available.

No description available.

This thesis provides a syntactic and semantic analysis of bare arguments and classifiers in Thai as well as accounting for its nominal word order. Adopting the Nominal Mapping Parameter (Chierchia 1998), it is argued that Thai nouns are names of kinds. Kinds are of type <s,e>, which are allowed to appear without overt determiners in argument position. For this reason, Thai nouns cannot directly combine with a quantifier without the help of a classifier. The study shows that Thai arguments behave like English bare arguments (bare plurals and mass nouns) in that they exhibit scopelessness and can be interpreted with different meanings such as weak indefinite, generic and kind interpretations. Unlike English bare arguments, the Thai counterparts may also have a definite interpretation. This is because Thai lacks an overt definite determiner. In addition, the thesis provides a unified analysis for the occurrence of Thai classifiers in different contexts. It is assumed that a classifier occurs in a quantified context to provide a portion of a kind (Krifka 1995, Chierchia 1998). The thesis further proposes that a classifier occurs in a non-quantified context where there is no overt numeral when the noun phrase is specific. A specific noun phrase includes those appearing with a demonstrative, the numeral ‘one’ or a modifier. As for the word order within the nominal domain, it is proposed that the noun, although merged at the bottom of the Specific Phrase underlyingly, always appears in the initial position to check an uninterpretable nominal feature in the Specific head.

This thesis provides a syntactic and semantic analysis of bare arguments and classifiers in Thai as well as accounting for its nominal word order. Adopting the Nominal Mapping Parameter (Chierchia 1998), it is argued that Thai nouns are names of kinds. Kinds are of type <s,e>, which are allowed to appear without overt determiners in argument position. For this reason, Thai nouns cannot directly combine with a quantifier without the help of a classifier. The study shows that Thai arguments behave like English bare arguments (bare plurals and mass nouns) in that they exhibit scopelessness and can be interpreted with different meanings such as weak indefinite, generic and kind interpretations. Unlike English bare arguments, the Thai counterparts may also have a definite interpretation. This is because Thai lacks an overt definite determiner. In addition, the thesis provides a unified analysis for the occurrence of Thai classifiers in different contexts. It is assumed that a classifier occurs in a quantified context to provide a portion of a kind (Krifka 1995, Chierchia 1998). The thesis further proposes that a classifier occurs in a non-quantified context where there is no overt numeral when the noun phrase is specific. A specific noun phrase includes those appearing with a demonstrative, the numeral ‘one’ or a modifier. As for the word order within the nominal domain, it is proposed that the noun, although merged at the bottom of the Specific Phrase underlyingly, always appears in the initial position to check an uninterpretable nominal feature in the Specific head.

No description available.

Digital multimedia content production and the amount of content present all over the world have exploded in the recent years. The consequences of this fact can be observed everywhere in many different forms, to exemplify, huge digital video archives of broadcasting companies, commercial image archives, virtual museums, etc. In order for these sources to be useful and accessible, this technological advance must be accompanied by the effective techniques of indexing and retrieval. The most effective way of indexing is the one providing a basis for retrieval in terms of semantic concepts, upon which ordinary users of multimedia databases base their queries. On the other hand, semantic classification of images using low-level features is a challenging problem. Combining experts with different classifier structures, trained by MPEG-7low-level color and texture descriptors, is examined as a solution alternative. For combining different classifiers and features, advanced decision mechanisms are proposed, which utilize basic expert combination strategies in different settings. Each of these decision mechanisms, namely Single Feature Combination (SFC), Multiple Feature Direct Combination (MFDC), and Multiple Feature Cascaded Combination (MFCC) enjoy significant classification performance improvements over single experts. Simulations are conducted on eight different visual semantic classes, resulting in accuracy improvements between 3.5-6.5%, when they are compared with the best performance of single expert systems.

This thesis proposes four novel geometric neurodynamical classifier models, namely GBAM, Lie-derivative, Lie-Poisson, and FAM, applied to breast cancer detection. All these models have been published in a paper and/or in a book form. All theoretical material of this thesis (Chapter 2) has been published in my monographs (see my publication list), as follows: 2.1 Tensorial Neurodynamics has been published in Natural Biodynamics (Chapters 3, 5 and 7), Geometrical Dynamics of Complex Systems; (Chapter 1 and Appendix), 2006) as well as Applied Differential Geometry:A Modern Introduction(Chapter 3) 2.2 GBAM Neurodynamical Classifier has been published in Natural Biodynamics (Chapter 7) and Neuro-Fuzzy Associative Machinery for Comprehensive Brain and Cognition Modelling (Chapter 3), as well as in the KES–Conference paper with the same title; 2.3 Lie-Derivative Neurodynamical Classifier has been published in Geometrical Dynamics of Complex Systems; (Chapter 1) and Applied Differential Geometry: A Modern Introduction (Chapter 3); 2.4 Lie-Poisson Neurodynamical Classifier has been published in Geometrical Dynamics of Complex Systems; (Chapter 1) and Applied Differential Geometry: A Modern Introduction (Chapter 3); 2.5 Fuzzy Associative Dynamical Classifier has been published in Neuro-Fuzzy Associative Machinery for Comprehensive Brain and Cognition Modelling (Chapter 4), as well as in the KES-Conference paper with the same title. Besides, Section 1.2 Artificial Neural Networks has been published in Natural Biodynamics (Chapter 7) and Neuro-Fuzzy Associative Machinery for Comprehensive Brain and Cognition Modelling (Chapter 3). Also, Sections 4.1. and 4.5. have partially been published in Neuro-Fuzzy Associative Machinery for Comprehensive Brain and Cognition Modelling (Chapters 3 and 4, respectively) and in the corresponding KES–Conference papers. A. The GBAM (generalized bidirectional associative memory) classifier is a neurodynamical, tensor-invariant classifier based on Riemannian geometry. The GBAM is a tensor-field system resembling a two-phase biological neural oscillator in which an excitatory neural field excites an inhibitory neural field, which reciprocally inhibits the excitatory one. This is a new generalization of Kosko’s BAM neural network, with a new biological (oscillatory, i.e., excitatory/inhibitory)interpretation. The model includes two nonlinearly-coupled (yet non-chaotic and Lyapunov stable) subsystems, activation dynamics and self-organized learning dynamics, including a symmetric synaptic 2-dimensional tensor-field, updated by differential Hebbian associative learning innovations. Biologically, the GBAM describes interacting excitatory and inhibitory populations of neurons found in the cerebellum, olfactory cortex, and neocortex, all representing the basic mechanisms for the generation of oscillating (EEG-monitored) activity in the brain. B. Lie-derivative neurodynamical classifier is an associative-memory, tensor-invariant neuro-classifier, based on the Lie-derivative operator from geometry of smooth manifolds. C. Lie-Poisson neurodynamical classifier is an associative-memory, tensor-invariant neuro-classifier based on the Lie-Poisson bracket from the generalized symplectic geometry. D. The FAM-matrix (fuzzy associative memory) dynamical classifier is a fuzzy-logic classifier based on a FAM-matrix (fuzzy phase-plane). All models are formulated and simulated in Mathematica computer algebra system. All models are applied to breast cancer detection, using the database from the University of Wisconsin and Mammography database. Classification results outperformed those obtained with standard MLP trained with backpropagation algorithm. / Thesis (Ph.D.) -- University of Adelaide, School of Mathematical Sciences, 2008