Dat Tran

Followers

Following

Public Views

Interests

Uploads

Papers by Dat Tran

Fuzzy Gaussian mixture models for speaker recognition

5th International Conference on Spoken Language Processing (ICSLP 1998)

The Gaussian mixture model (GMM) is an important application of statistical clustering to speaker... more The Gaussian mixture model (GMM) is an important application of statistical clustering to speaker recognition. A number of prototypes are generated from the training feature vectors by representing the feature space as a mixture of Gaussian distributions. Each prototype consists of a model parameter set including mean vector, covariance matrix and mixture weight. In fuzzy clustering, the fuzzy c-means (FCM) method is the most widely used. Model parameters in each prototype include fuzzy mean vector and fuzzy covariance matrix. Both the GMM and the FCM methods have similar characteristics: using iterative optimisation algorithms, feature vectors can belong to more than one class, and degrees of belonging of a vector across classes sum to one. From these similarities, a FCM-based generalisation to the GMM called the fuzzy GMM (FGMM) is proposed in this paper. Fuzzy mixture weights are introduced by redefining the distances in the FCM functionals. The FGMM algorithm and its use in speaker recognition are considered The experimental results show that with a suitable degree offuzziness, the FGMMs are more effective than the GMMs in tests on 16 speakers using the T/46 database and on 108 speakers using the ANDOSL database.

Download

Fuzzy hidden Markov models for speech and speaker recognition

18th International Conference of the North American Fuzzy Information Processing Society - NAFIPS (Cat. No.99TH8397)

This paper proposes a Frzzy approach to the hidden Markov model (HMM) method called the futzr HMM... more This paper proposes a Frzzy approach to the hidden Markov model (HMM) method called the futzr HMM for speech and speaker recognition The fuzzy HMM algorithm is regarded as an application of the fuzzy expectation-maximisation (EM) algorithm to the Baum-Welch algorithm in the HMM. Speech and speaker recognition experiments using the Texas Instruments p 4 6) speech data corpus show better results for fuuy HMMs compared with conventioml HMMs.

Download

A proposed likelihood transformation for speaker verification

2000 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings (Cat. No.00CH37100)

Most of current normalisation methods for speaker verifi-cation are based on the ratio of the cla... more Most of current normalisation methods for speaker verifi-cation are based on the ratio of the claimed speaker&amp;amp;amp;#x27;s and impostors&amp;amp;amp;#x27; likelihood functions. As an extension, we pro-pose alternative normalisation methods, which are based on the ratio of functions of the claimed speaker&amp;amp;amp;#x27;s ...

Tablet PC Applications in an Academic Environment

ICTACS 2006, 2006

Tablet PCs are a new generation of notebook computers which provide multimodal input options of p... more Tablet PCs are a new generation of notebook computers which provide multimodal input options of pen, voice and keyboard. Recently, these portable and flexible Tablet PCs have attracted attention as a potential tool in academic environments. This paper reviews the current use of Tablet PCs in teaching computer science and software engineering courses, presenting lectures and papers, and creating peer-review comments. The paper also presents applications of Tablet PCs in teaching and research at our university, the University of Canberra, Australia. These applications include marking assignments and reports, and developing signature verification applications.

Download

Dynamic Scheduling Using Multiagent Architecture

Lecture Notes in Computer Science, 2004

A dynamic approach to truck scheduling using an intelligent distributed multi-agent architecture ... more A dynamic approach to truck scheduling using an intelligent distributed multi-agent architecture is presented. This paper presents an inherently distributed problem, the dynamic truck scheduling problem which has motivated the current study on an intelligent multi-agent architecture for optimal real time solutions. The .Net environment is critiqued for the proposed multi-agent architecture. A design and implementation of a prototype system is presented and some test results discussed. Possible future extensions to the system to cater for communication through hand-held devices for mobile agents required in the truck scheduling problem are also summarized.

Download

Generalised Fuzzy Hidden Markov Models for Speech Recognition

Lecture Notes in Computer Science, 2002

A generalised fuzzy approach to statistical modelling techniques for speech recognition is propos... more A generalised fuzzy approach to statistical modelling techniques for speech recognition is proposed in this paper. Fuzzy C-means (FCM) and fuzzy entropy (FE) techniques are combined into a generalised fuzzy technique and applied to hidden Markov models (HMMs). A more robust version of the above fuzzy technique based on the noise clustering (NC) method is also proposed. Experimental results were performed on the TI46 speech data corpus. A significant result for isolatedword recognition performed on a highly confusable vocabulary consisting of the nine English E-set words is that, a 33.8% recognition error rate for the HMM-based system was reduced to 30.5%, 29.9%, 29.8% and 27.8%, respectively, by using the FCM-HMM, the FE-HMM, the NC-FE-HMM, and the NC-FCM-HMM-based systems.

Download

Adaptable Term Weighting Framework for Text Classification

Lecture Notes in Computer Science, 2011

In text classification, term frequency and term co-occurrence factors are dominantly used in weig... more

Grammatical Dependency-Based Relations for Term Weighting in Text Classification

Lecture Notes in Computer Science, 2011

Term frequency and term co-occurrence are currently used to estimate term weightings in a documen... more

Fuzzy Subspace Hidden Markov Models for Pattern Recognition

2009 IEEE-RIVF International Conference on Computing and Communication Technologies, 2009

This paper presents a novel fuzzy subspace-based approach to hidden Markov model. Features extrac... more This paper presents a novel fuzzy subspace-based approach to hidden Markov model. Features extracted from patterns are considered as feature vectors in a multi-dimensional feature space. Current hidden Markov modeling techniques treat features equally, however this assumption may not be true. We propose to consider subspaces in the feature space and assign a weight to each feature to determine the contribution of that feature in different subspaces to modeling and recognizing patterns. Weights can be computed if a learning estimation method such as maximum likelihood is given. Experimental results in network intrusion detection based on the proposed approach show promising results.

Download

Filtering Spam Email with Flexible Preprocessors

Lecture Notes in Electrical Engineering, 2008

ABSTRACT

Classification of gender and face based on gradient faces

3rd European Workshop on Visual Information Processing, 2011

This paper presents a new method for solving face gender identification and face classification p... more This paper presents a new method for solving face gender identification and face classification problems. The proposed method uses gradient features for feature extraction and support vector machine for classification. Experiments for the proposed method have been conducted on two public data sets CalTech and AT&T. The results show that the proposed method could improve the classification rates.

Automatic classification of speaker characteristics

International Conference on Communications and Electronics 2010, 2010

An automatic voice-based classification system of speaker characteristics including age, gender a... more An automatic voice-based classification system of speaker characteristics including age, gender and accent is presented in this paper. Speakers are grouped according to their characteristics and their speech features are then extracted to train speaker group models using different classification techniques. Finally fusion of classification results for those speaker groups is performed to obtain results for each speaker characteristic. The ANDOSL Australian speech database consisting of 108 speakers and 21600 long utterances was used for system evaluation. Experiments showed high performance for the proposed classification of speaker characteristics. Keywords-component; speaker characteristics; speech processing; vector quantization; Gaussian mixture model; support vector machine. I. INTRODUCTION Speaker characteristics can be divided in to relatively stable characteristics and transient characteristics. Stable speaker characteristics comprise physiological and anatomical factors such as gender and age. Transient speaker characteristics comprise stress and emotional state. Stable speaker characteristics are easier to recognise [1]. The most important stable speaker characteristics will be mentioned below. Classifying speaker characteristics is an important task in Dialog Systems, Speech Synthesis, Forensics, Language Learning, Assessment Systems, and Speaker Recognition Systems [2]. In Human-Computer Interaction applications, the interaction between users and computers taking place at the speech-driven user interface. For example, Spoken Dialogs Systems provide services in domains of finance, travel, scheduling, tutoring, or weather. The systems need to gather automatically information from the user in order to provide timely and relevant services. Most telephone-based services today use spoken dialog systems to either route calls to the appropriate agent or even handle the complete service by an automatic system. Another example of Human-Computer Interaction application is Computer-aided Learning and Assessment systems. The systems provide interactive recording and playback of user's input speech, feedback regarding acoustic speech features, recognizing the input, and interpreting interaction to act as a conversation partner. Besides customizing to the native language of the language learner, learning systems may have to be tailored towards particular accents, for example the E-Language Learning System program between the U.S. Department of Education

Download

Fuzzy feature weighting techniques for vector quantisation

International Conference on Fuzzy Systems, 2010

ABSTRACT Vector quantization (VQ) is a simple but effective modelling technique in pattern recogn... more ABSTRACT Vector quantization (VQ) is a simple but effective modelling technique in pattern recognition. VQ employs a clustering technique to convert a feature vector set in to a cluster center set to model the feature vector set. Some clustering techniques have been applied to improve VQ. However VQ is not always effective because data features are treated equally although their importance may not be the same. Some automated feature weighting techniques have been proposed to overcome this drawback. This paper reviews those weighting techniques and proposes a general scheme for selecting any pair of clustering and feature weighting techniques to form a fuzzy feature weighting-based VQ modelling technique. Besides the current techniques, a number of new feature weighting-based VQ techniques is proposed and their evaluations are also presented.

State mixture modelling applied to speech recognition

Pattern Recognition Letters, 1999

In state mixture modelling (SMM), the temporal structure of the observation sequences is represen... more In state mixture modelling (SMM), the temporal structure of the observation sequences is represented by the state joint probability distribution where mixtures of states are considered. This technique is considered in an iterative scheme via maximum likelihood estimation. A fuzzy estimation approach is also introduced to cooperate with the SMM model. This new approach not only saves calculations from 2x (HMM direct calculation) and x 2 (Forward± backward algorithm) to just only 2NT calculations, but also achieves a better recognition result.

Download

The Big Australian Speech Corpus (The Big ASC)

13th Australasian International Conference on Speech Science and Technology, 2010

Under an ARC Linkage Infrastructure, Equipment and Facilities (LIEF) grant, speech science and te... more Under an ARC Linkage Infrastructure, Equipment and Facilities (LIEF) grant, speech science and technology experts from across Australia have joined forces to organise the recording of audio-visual (AV) speech data from representative speakers of Australian English in all capital cities and some regional centres. The Big Australian Speech Corpus (the Big ASC) will provide a standard recording setup and a collaboratively-designed elicitation protocol to create a corpus of AV speech data incorporating annotations and ...

Download

Remote multimodal biometric authentication using bit priority-based fragile watermarking

2008 19th International Conference on Pattern Recognition, 2008

We propose a new remote multimodal biometric authentication framework based on fragile watermarki... more We propose a new remote multimodal biometric authentication framework based on fragile watermarking for transferring multi-biometrics over networks to server for authentication. A facial image is used as a container to embed other numeric biometrics features. The proposed framework enhances security and reduces bandwidths. In order to reduce error rates from embedding numeric information, we also propose a new method to determine bit priority level in a bit sequence representing the numerical information to be embedded and combine with the current amplitude modulation watermarking method.

Download

Dat Tran

Uploads

Papers by Dat Tran

Log In