Linguistic constraints for large vocabulary speech recognition.

by Roger H.Y. Leung.Thesis (M.Phil.)--Chinese University of Hong Kong, 1999.Includes bibliographical references (leaves 79-84).Abstracts in English and Chinese.ABSTRACT --- p.IKeywords: --- p.IACKNOWLEDGEMENTS --- p.IIITABLE OF CONTENTS: --- p.IVTable of Figures: --- p.VITable of Tables: --- p.VIIChapter CHAPTER 1 --- INTRODUCTION --- p.1Chapter 1.1 --- Languages in the World --- p.2Chapter 1.2 --- Problems of Chinese Speech Recognition --- p.3Chapter 1.2.1 --- Unlimited word size: --- p.3Chapter 1.2.2 --- Too many Homophones: --- p.3Chapter 1.2.3 --- Difference between spoken and written Chinese: --- p.3Chapter 1.2.4 --- Word Segmentation Problem: --- p.4Chapter 1.3 --- Different types of knowledge --- p.5Chapter 1.4 --- Chapter Conclusion --- p.6Chapter CHAPTER 2 --- FOUNDATIONS --- p.7Chapter 2.1 --- Chinese Phonology and Language Properties --- p.7Chapter 2.1.1 --- Basic Syllable Structure --- p.7Chapter 2.2 --- Acoustic Models --- p.9Chapter 2.2.1 --- Acoustic Unit --- p.9Chapter 2.2.2 --- Hidden Markov Model (HMM) --- p.9Chapter 2.3 --- Search Algorithm --- p.11Chapter 2.4 --- Statistical Language Models --- p.12Chapter 2.4.1 --- Context-Independent Language Model --- p.12Chapter 2.4.2 --- Word-Pair Language Model --- p.13Chapter 2.4.3 --- N-gram Language Model --- p.13Chapter 2.4.4 --- Backoff n-gram --- p.14Chapter 2.5 --- Smoothing for Language Model --- p.16Chapter CHAPTER 3 --- LEXICAL ACCESS --- p.18Chapter 3.1 --- Introduction --- p.18Chapter 3.2 --- Motivation： Phonological and lexical constraints --- p.20Chapter 3.3 --- Broad Classes Representation --- p.22Chapter 3.4 --- Broad Classes Statistic Measures --- p.25Chapter 3.5 --- Broad Classes Frequency Normalization --- p.26Chapter 3.6 --- Broad Classes Analysis --- p.27Chapter 3.7 --- Isolated Word Speech Recognizer using Broad Classes --- p.33Chapter 3.8 --- Chapter Conclusion --- p.34Chapter CHAPTER 4 --- CHARACTER AND WORD LANGUAGE MODEL --- p.35Chapter 4.1 --- Introduction --- p.35Chapter 4.2 --- Motivation --- p.36Chapter 4.2.1 --- Perplexity --- p.36Chapter 4.3 --- Call Home Mandarin corpus --- p.38Chapter 4.3.1 --- Acoustic Data --- p.38Chapter 4.3.2 --- Transcription Texts --- p.39Chapter 4.4 --- Methodology: Building Language Model --- p.41Chapter 4.5 --- Character Level Language Model --- p.45Chapter 4.6 --- Word Level Language Model --- p.48Chapter 4.7 --- Comparison of Character level and Word level Language Model --- p.50Chapter 4.8 --- Interpolated Language Model --- p.54Chapter 4.8.1 --- Methodology --- p.54Chapter 4.8.2 --- Experiment Results --- p.55Chapter 4.9 --- Chapter Conclusion --- p.56Chapter CHAPTER 5 --- N-GRAM SMOOTHING --- p.57Chapter 5.1 --- Introduction --- p.57Chapter 5.2 --- Motivation --- p.58Chapter 5.3 --- Mathematical Representation --- p.59Chapter 5.4 --- Methodology: Smoothing techniques --- p.61Chapter 5.4.1 --- Add-one Smoothing --- p.62Chapter 5.4.2 --- Witten-Bell Discounting --- p.64Chapter 5.4.3 --- Good Turing Discounting --- p.66Chapter 5.4.4 --- Absolute and Linear Discounting --- p.68Chapter 5.5 --- Comparison of Different Discount Methods --- p.70Chapter 5.6 --- Continuous Word Speech Recognizer --- p.71Chapter 5.6.1 --- Experiment Setup --- p.71Chapter 5.6.2 --- Experiment Results: --- p.72Chapter 5.7 --- Chapter Conclusion --- p.74Chapter CHAPTER 6 --- SUMMARY AND CONCLUSIONS --- p.75Chapter 6.1 --- Summary --- p.75Chapter 6.2 --- Further Work --- p.77Chapter 6.3 --- Conclusion --- p.78REFERENCE --- p.7