decide which contexts are similar and can share parameters. A key feature of this approach is that it allows the construction of models which are dependent upon contextual effects occurring across word boundaries. The use of cross word context dependent models presents problems for conventional decoders. The second part of the thesis therefore presents a new decoder design which is capable of using these models efficiently. The decoder is suitable for use with very large vocabularies and long span language models. It is also capable of generating a lattice of word hypotheses with little computational overhead. These lattices can be used to constrain further decoding, allowing efficient use of complex acoustic and language models. The effectiveness of these techniques has been assessed on a variety of large vocabulary continuous speech recognition tasks and results are presented which analyse performance in terms of computational complexity and recognition accuracy. The experiments dem

This paper investigates the use of Gaussian Selection (GS) to reduce the state likelihood computation in HMM-based systems. These likelihood calculations contribute significantly (30 to 70%) to the computational load. Previously, it has been reported that when GS is used on large systems the recognition accuracy tends to degrade above a \Theta3 reduction in likelihood computation. To explain this degradation, this paper investigates the trade-offs necessary between achieving good state likelihoods and low computation. In addition, the problem of unseen states in a cluster is examined. It is shown that further improvements are possible. For example, using a different assignment measure, with a constraint on the number of components per state per cluster, enabled the recognition accuracy on a 5k speaker-independent task to be maintained up to a \Theta5 reduction in likelihood computation. 1. INTRODUCTION In recent years, high accuracy large vocabulary continuous speech recognition sys...

: : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : xiii 1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1 1.1 The Problem: Too Many Parameters : : : : : : : : : : : : : : : : : : : : : : 3 1.2 Proposed Solution: It Is Time to Share More! : : : : : : : : : : : : : : : : : 4 1.3 Thesis Summary and Outline : : : : : : : : : : : : : : : : : : : : : : : : : : 6 2 Review of Acoustic Modeling Using Hidden Markov Model : : : : : : : 9 2.1 Speech Characteristics : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 9 2.2 Selection of Input Speech Space and Speech Model : : : : : : : : : : : : : : 10 2.2.1 Cepstral Input : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 10 2.2.2 Hidden Markov Model : : : : : : : : : : : : : : : : : : : : : : : : : : 11 2.2.3 Our Choice of HMM for Acoustic Modeling : : : : : : : : : : : : : : 14 2.3 Speech Unit to Model : : : : : : : : : : : : : : : : : : : : : : : : : : ...

. This paper reports experiments measuring the e#ects of #- nite precision arithmetic in the range of 4 to 16 bits on three particular pattern-recognition algorithms: an optical character recognizer #1#, a pen-based character recognizer #2#, and a speech recognizer #3#. The measurements shows that large portions of these algorithms can be implemented with 8-bit arithmetic #e.g. using Intel's MMX TM instruction set# while incurring only a negligible loss in recognition accuracy. 1 Introduction Three particular pattern recognition algorithms # an optical character recognizer #1#, a pen-based handwriting recognizer #2#, and a speech recognizer #3# # are experimentally analyzed with respect to the degradation caused by #nite precision arithmetic. The recognition accuracy or error rate is plotted as a function of the arithmetic precision ranging from 4 to 16 bits and is compared to the original 32-bit #oating-point implementation. This analysis is motivated by the introduction of...

This paper investigates the use of Gaussian Selection (GS) to reduce the state likelihood computation in HMM-based systems. These likelihood calculations contribute significantly (30 to 70%) to the computational load. Previously, it has been reported that when GS is used on large systems the recognition accuracy tends to degrade above a #3 reduction in likelihood computation. To explain this degradation, this paper investigates the trade-offs necessary between achieving good state likelihoods and low computation. In addition, the problem of unseen states in a cluster is examined. It is shown that further improvements are possible. For example, using a different assignment measure, with a constraint on the number of components per state per cluster, enabled the recognition accuracy on a 5k speaker-independent task to be maintained up to a #5 reduction in likelihood computation.

This patent concerns the traditional problem encountered in the syntactic Pattern Recognition (PR) of strings or sequences. The primary investigator 1 involved in this work is a Full Professor at Carleton University in Ottawa, Canada, and is a Fellow of the IEEE. The primary problem solved by the invention involves determining the string or sequence that is most similar to a sequence presented to the system. The search could be initiated by presenting, to the system, a noisy or inexact version of a string contained in memory-for example, at a web-site or in the library or database. The invention will yield the closest string/sequence by searching the dictionary of possible words using a newly invented AIbased strategy. The core of this invention is this search strategy, called the Clustered Beam Search. Experiments have been done to show the benefits of the CBS over the current state-ofthe-art, and the results demonstrate an unbelievably marked improvement (sometimes as high as 90%) for large libraries and databases. The solution provided by the invention would be applicable in numerous areas including: Inexact or proximity searching on the Internet, keyword-based search in libraries and databases, spelling correction, speech and character recognition (including optical character recognition), and the processing of biological sequences, for example, in human genome projects. These applications are briefly described below.