The Application of Hidden Markov Models in Speech

The Application of Hidden Markov Models in Speech Recognition Author：Mark Gales 1 and Steve Young 2 Published： 21 Feb 2008 Subjects：Speech/audio/image/video compression

Outline Introduction Architecture of an HMM-Based Recogniser HMM Structure Refinements references

CHAPTER 1: Introduction

There is no data like more data. Recognition word error rate vs. the amount of training hours for illustrative purposes only. This figure illustrates how modern speech recognition systems can benefit from increased training data.

Automatic continuous speech recognition (CSR) has many potential applications including command control, dictation, transcription of recorded speech, searching audio documents and interactive spoken dialogues. in the last decade or more, the detailed modelling techniques developed within this framework have evolved to a state of considerable sophistication. Since speech has temporal structure and can be encoded as a sequence of spectral vectors spanning the audio frequency range, the hidden Markov model (HMM) provides a natural framework for constructing such models [13].

CHAPTER 2: Architecture of an HMM-Based Recogniser

Architecture of a HMM-based Recogniser Bayes’ Rule Acoustic Models Language Model

HMM-based phone model HMM Acoustic Models

Formation of tied-state phone models

N-gram Language Models

CHAPTER 3: HMM Structure Refinements

Dynamic Bayesian Networks In Architecture of an HMM-Based Recogniser, the HMM was describedas a generative model which for a typical phone has three emitting

Gaussian Mixture Models

Covariance Modelling Structured Covariance Matrices Structured Precision Matrices

references 使用中 The Application of Hidden Markov Modelsin Speech Recognition A Historical Perspective of Speech Recognition Automatic Speech Recognition – A Brief History of the Technology Development Bayesian Networks