Spectral coding methods for speech compression and speaker identification

This thesis investigates aspects of encoding the speech spectrum at low bit rates, with extensions to the effect of such coding on automatic speaker identification. Vector quantization (VQ) is a technique for jointly quantizing a block of samples at once, in order to reduce the bit rate of a coding system. The major drawback in using VQ is the complexity of the encoder. Recent research has indicated the potential applicability of the VQ method to speech when product code vector quantization (PCVQ) techniques are utilized. The focus of this research is the efficient representation, calculation and utilization of the speech model as stored in the PCVQ codebook. In this thesis, several VQ approaches are evaluated, and the efficacy of two training algorithms is compared experimentally. It is then shown that these productcode vector quantization algorithms may be augmented with lossless compression algorithms, thus yielding an improved overall compression rate. An approach using a statistical model for the vector codebook indices for subsequent lossless compression is introduced. This coupling of lossy compression and lossless compression enables further compression gain. It is demonstrated that this approach is able to reduce the bit rate requirement from the current 24 bits per 20 millisecond frame to below 20, using a standard spectral distortion metric for comparison. Several fast-search VQ methods for use in speech spectrum coding have been evaluated. The usefulness of fast-search algorithms is highly dependent upon the source characteristics and, although previous research has been undertaken for coding of images using VQ codebooks trained with the source samples directly, the product-code structured codebooks for speech spectrum quantization place new constraints on the search methodology. The second major focus of the research is an investigation of the effect of lowrate spectral compression methods on the task of automatic speaker identification. The motivation for this aspect of the research arose from a need to simultaneously preserve the speech quality and intelligibility and to provide for machine-based automatic speaker recognition using the compressed speech. This is important because there are several emerging applications of speaker identification where compressed speech is involved. Examples include mobile communications where the speech has been highly compressed, or where a database of speech material has been assembled and stored in compressed form. Although these two application areas have the same objective - that of maximizing the identification rate - the starting points are quite different. On the one hand, the speech material used for training the identification algorithm may or may not be available in compressed form. On the other hand, the new test material on which identification is to be based may only be available in compressed form. Using the spectral parameters which have been stored in compressed form, two main classes of speaker identification algorithm are examined. Some studies have been conducted in the past on bandwidth-limited speaker identification, but the use of short-term spectral compression deserves separate investigation. Combining the major aspects of the research, some important design guidelines for the construction of an identification model when based on the use of compressed speech are put forward.

Stationary processes for object detection and non-rigid structure-from-motion

Stationary processes are random variables whose value is a signal and whose distribution is invariant to translation in the domain of the signal. They are intimately connected to convolution, and therefore to the Fourier transform, since the covariance matrix of a stationary process is a Toeplitz matrix, and Toeplitz matrices are the expression of convolution as a linear operator. This thesis utilises this connection in the study of i) efficient training algorithms for object detection and ii) trajectory-based non-rigid structure-from-motion.

Correlating eye gaze direction, depth and vehicle information on an interactive map for driver training

The over represented number of novice drivers involved in crashes is alarming. Driver training is one of the interventions aimed at mitigating the number of crashes that involve young drivers. Experienced drivers have better hazard perception ability compared to inexperienced drivers. Eye gaze patterns have been found to be an indicator of the driver's competency level. The aim of this paper is to develop an in-vehicle system which correlates information about the driver's gaze and vehicle dynamics, which is then used to assist driver trainers in assessing driving competency. This system allows visualization of the complete driving manoeuvre data on interactive maps. It uses an eye tracker and perspective projection algorithms to compute the depth of gaze and plots it on Google maps. This interactive map also features the trajectory of the vehicle and turn indicator usage. This system allows efficient and user friendly analysis of the driving task. It can be used by driver trainers and trainees to understand objectively the risks encountered during driving manoeuvres. This paper presents a prototype that plots the driver's eye gaze depth and direction on an interactive map along with the vehicle dynamics information. This prototype will be used in future to study the difference in gaze patterns in novice and experienced drivers prior to a certain manoeuvre.

Hierarchical fuzzy logic traffic control at a road junction using genetic algorithms

Traffic control at a road junction by a complex fuzzy logic controller is investigated. The increase in the complexity of junction means more number of input variables must be taken into account, which will increase the number of fuzzy rules in the system. A hierarchical fuzzy logic controller is introduced to reduce the number of rules. Besides, the increase in the complexity of the controller makes formulation of the fuzzy rules difficult. A genetic algorithm based off-line leaning algorithm is employed to generate the fuzzy rules. The learning algorithm uses constant flow-rates as training sets. The system is tested by both constant and time-varying flow-rates. Simulation results show that the proposed controller produces lower average delay than a fixed-time controller does under various traffic conditions.

Selected new training documents to update user profile

Relevance Feedback (RF) has been proven very effective for improving retrieval accuracy. Adaptive information filtering (AIF) technology has benefited from the improvements achieved in all the tasks involved over the last decades. A difficult problem in AIF has been how to update the system with new feedback efficiently and effectively. In current feedback methods, the updating processes focus on updating system parameters. In this paper, we developed a new approach, the Adaptive Relevance Features Discovery (ARFD). It automatically updates the system's knowledge based on a sliding window over positive and negative feedback to solve a nonmonotonic problem efficiently. Some of the new training documents will be selected using the knowledge that the system currently obtained. Then, specific features will be extracted from selected training documents. Different methods have been used to merge and revise the weights of features in a vector space. The new model is designed for Relevance Features Discovery (RFD), a pattern mining based approach, which uses negative relevance feedback to improve the quality of extracted features from positive feedback. Learning algorithms are also proposed to implement this approach on Reuters Corpus Volume 1 and TREC topics. Experiments show that the proposed approach can work efficiently and achieves the encouragement performance.

Exponentiated gradient algorithms for conditional random fields and max-margin Markov Networks

Log-linear and maximum-margin models are two commonly-used methods in supervised machine learning, and are frequently used in structured prediction problems. Efficient learning of parameters in these models is therefore an important problem, and becomes a key factor when learning from very large data sets. This paper describes exponentiated gradient (EG) algorithms for training such models, where EG updates are applied to the convex dual of either the log-linear or max-margin objective function; the dual in both the log-linear and max-margin cases corresponds to minimizing a convex function with simplex constraints. We study both batch and online variants of the algorithm, and provide rates of convergence for both cases. In the max-margin case, O(1/ε) EG updates are required to reach a given accuracy ε in the dual; in contrast, for log-linear models only O(log(1/ε)) updates are required. For both the max-margin and log-linear cases, our bounds suggest that the online EG algorithm requires a factor of n less computation to reach a desired accuracy than the batch EG algorithm, where n is the number of training examples. Our experiments confirm that the online algorithms are much faster than the batch algorithms in practice. We describe how the EG updates factor in a convenient way for structured prediction problems, allowing the algorithms to be efficiently applied to problems such as sequence learning or natural language parsing. We perform extensive evaluation of the algorithms, comparing them to L-BFGS and stochastic gradient descent for log-linear models, and to SVM-Struct for max-margin models. The algorithms are applied to a multi-class problem as well as to a more complex large-scale parsing task. In all these settings, the EG algorithms presented here outperform the other methods.

Exponentiated gradient algorithms for large-margin structured classification

We consider the problem of structured classification, where the task is to predict a label y from an input x, and y has meaningful internal structure. Our framework includes supervised training of Markov random fields and weighted context-free grammars as special cases. We describe an algorithm that solves the large-margin optimization problem defined in [12], using an exponential-family (Gibbs distribution) representation of structured objects. The algorithm is efficient—even in cases where the number of labels y is exponential in size—provided that certain expectations under Gibbs distributions can be calculated efficiently. The method for structured labels relies on a more general result, specifically the application of exponentiated gradient updates [7, 8] to quadratic programs.

Part 2. MOO methods for multidisciplinary design using parallel evolutionary algorithms, game theory and hierarchical topology. (Part 2.) Numerical aspects and implementation of model test cases

These lecture notes describe the use and implementation of a framework in which mathematical as well as engineering optimisation problems can be analysed. The foundations of the framework and algorithms described -Hierarchical Asynchronous Parallel Evolutionary Algorithms (HAPEAs) - lie upon traditional evolution strategies and incorporate the concepts of a multi-objective optimisation, hierarchical topology, asynchronous evaluation of candidate solutions , parallel computing and game strategies. In a step by step approach, the numerical implementation of EAs and HAPEAs for solving multi criteria optimisation problems is conducted providing the reader with the knowledge to reproduce these hand on training in his – her- academic or industrial environment.

Centroid training to achieve effective text classification

Traditional text classification technology based on machine learning and data mining techniques has made a big progress. However, it is still a big problem on how to draw an exact decision boundary between relevant and irrelevant objects in binary classification due to much uncertainty produced in the process of the traditional algorithms. The proposed model CTTC (Centroid Training for Text Classification) aims to build an uncertainty boundary to absorb as many indeterminate objects as possible so as to elevate the certainty of the relevant and irrelevant groups through the centroid clustering and training process. The clustering starts from the two training subsets labelled as relevant or irrelevant respectively to create two principal centroid vectors by which all the training samples are further separated into three groups: POS, NEG and BND, with all the indeterminate objects absorbed into the uncertain decision boundary BND. Two pairs of centroid vectors are proposed to be trained and optimized through the subsequent iterative multi-learning process, all of which are proposed to collaboratively help predict the polarities of the incoming objects thereafter. For the assessment of the proposed model, F1 and Accuracy have been chosen as the key evaluation measures. We stress the F1 measure because it can display the overall performance improvement of the final classifier better than Accuracy. A large number of experiments have been completed using the proposed model on the Reuters Corpus Volume 1 (RCV1) which is important standard dataset in the field. The experiment results show that the proposed model has significantly improved the binary text classification performance in both F1 and Accuracy compared with three other influential baseline models.

A Best Practice Process for Fleet Safety Training

This paper outlines a process for fleet safety training based on research and management development programmes undertaken at the University of Huddersfield in the UK (www.hud.ac.uk/sas/trans/transnews.htm) and CARRS-Q in Australia (www.carrsq.qut.edu.au/staff/Murray.jsp) over the past 10 years.