This landmark book offers a balanced discussion of both the mathematical theory of digital speech signal processing and critical contemporary applications. Authors John R. Deller, John H. L. Hansen and John G. Proakis provide a comprehensive view of all major modern speech processing areas: speech production physiology and modeling, signal analysis techniques, coding, enhancement, quality assessment, and recognition. You will learn the principles needed to understand advanced technologies in speech processing, from speech coding for communications systems to biomedical applications of speech analysis and recognition.

Ideal for self-study or as a course text, this far-reaching reference book offers an extensive historical context for concepts under discussion, end-of-chapter problems, and practical algorithms.

Table of Contents

Preface to the IEEE Edition

Preface

Acronyms and Abbreviations

Signal Processing Background

Propaedeutic

Preamble

• The Purpose of Chapter 1
• Please Read This Note on Notation
• For People Who Never Read Chapter 1 (and Those Who Do)

Review of DSP Concepts and Notation

• "Normalized Time and Frequency"
• Singularity Signals
• Energy and Power Signals
• Transforms and a Few Related Concepts
• Windows and Frames
• Discrete-Time Systems
• Minimum, Maximum, and Mixed-Phase Signals and Systems

Review of Probability and Stochastic Processes

• Probability Spaces
• Random Variables
• Random Processes
• Vector-Valued Random Processes

Topics in Statistical Pattern Recognition

• Distance Measures
• The Euclidean Metric and "Prewhitening" of Features
• Maximum Likelihood Classification
• Feature Selection and Probablistic Separability Measures
• Clustering Algorithms

Information and Entropy

• Definitions
• Random Sources
• Entropy Concepts in Pattern Recognition

Phasors and Steady-State Solutions

Onward to Speech Processing

Problems

Appendices: Supplemental Bibliography

Example Textbooks on Digital Signal Processing

Example Textbooks on Stochastic Processes

Example Textbooks on Statistical Pattern Recognition

Example Textbooks on Information Theory

Other Resources on Speech Processing

• Textbooks
• Edited Paper Collections
• Journals
• Conference Proceedings

Example Textbooks on Speech and Hearing Sciences

Other Resources on Artificial Neural Networks

• Textbooks and Monographs
• Journals
• Conference Proceedings

Speech Production and Modeling

Fundamentals of Speech Science

Preamble

Speech Communication

Anatomy and Physiology of the Speech Production System

• Anatomy
• The Role of the Vocal Tract and Some Elementary Acoustical Analysis
• Excitation of the Speech System and the Physiology of Voicing

Phonemics and Phonetics

• Phonemes Versus Phones
• Phonemic and Phonetic Transcription
• Phonemic and Phonetic Classification
• Prosodic Features and Coarticulation

Conclusions

Problems

Modeling Speech Production

Preamble

Acoustic Theory of Speech Production

• History
• Sound Propagation
• Source Excitation Model
• Vocal-Tract Modeling
• Models for Nasals and Fricatives

Discrete-Time Modeling

• General Discrete-Time Speech Model
• A Discrete-Time Filter Model for Speech Production
• Other Speech Models

Conclusions

Problems

Single Lossless Tube Analysis

• Open and Closed Terminations
• Impedance Analysis, T-Network, and Two-Port Network

Two-Tube Lossless Model of the Vocal Tract

Fast Discrete-Time Transfer Function Calculation

Analysis Techniques

Short-Term Processing of Speech

Introduction

Short-Term Measures from Long-Term Concepts

• Motivation
• "Frames" of Speech
• Approach 1 to the Derivation of a Short-Term Feature and Its Two Computational Forms
• Approach 2 to the Derivation of a Short-Term Feature and Its Two Computational Forms
• On the Role of "1/N" and Related Issues

Example Short-Term Features and Applications

• Short-Term Estimates of Autocorrelation
• Average Magnitude Difference Function
• Zero Crossing Measure
• Short-Term Power and Energy Measures
• Short-Term Fourier Analysis

Conclusions

Problems

Linear Prediction Analysis

Preamble

Long-Term LP Analysis by System Identification

• The All-Pole Model
• Identification of the Model

How Good Is the LP Model?

• The "Ideal" and "Almost Ideal" Cases
• "Nonideal" Cases
• Summary and Further Discussion

Short-Term LP Analysis

• Autocorrelation Method
• Covariance Method
• Solution Methods
• Gain Computation
• A Distance Measure for LP Coefficients
• Preemphasis of the Speech Waveform

Alternative Representations of the LP Coefficients

• The Line Spectrum Pair
• Cepstral Parameters

Applications of LP in Speech Analysis

• Pitch Estimation
• Formant Estimation and Glottal Waveform Deconvolution

Conclusions

Problems

Proof of Theorem 5.1

The Orthogonality Principle

Cepstral Analysis

Introduction

"Real" Cepstrum

• Long-Term Real Cepstrum
• Short-Term Real Cepstrum
• Example Applications of the stRC to Speech Analysis and Recognition
• Other Forms and Variations on the stRC Parameters

Complex Cepstrum

• Long-Term Complex Cepstrum
• Short-Term Complex Cepstrum
• Example Application of the stCC to Speech Analysis
• Variations on the Complex Cepstrum

A Critical Analysis of the Cepstrum and Conclusions

Problems

Coding, Enhancement and Quality Assessment

Speech Coding and Synthesis

Introduction

Optimum Scalar and Vector Quantization

• Scalar Quantization
• Vector Quantization

Waveform Coding

• Introduction
• Time Domain Waveform Coding
• Frequency Domain Waveform Coding
• Vector Waveform Quantization

Vocoders

• The Channel Vocoder
• The Phase Vocoder
• The Cepstral (Homomorphic) Vocoder
• Formant Vocoders
• Linear Predictive Coding
• Vector Quantization of Model Parameters

Measuring the Quality of Speech Compression Techniques

Conclusions

Problems

Quadrature Mirror Filters

Speech Enhancement

Introduction

Classification of Speech Enhancement Methods

Short-Term Spectral Amplitude Techniques

• Introduction
• Spectral Subtraction
• Summary of Short-Term Spectral Magnitude Methods

Speech Modeling and Wiener Filtering

• Introduction
• Iterative Wiener Filtering
• Speech Enhancement and All-Pole Modeling
• Sequential Estimation via EM Theory
• Constrained Iterative Enhancement
• Further Refinements to Iterative Enhancement
• Summary of Speech Modeling and Wiener Filtering

Adaptive Noise Canceling

• Introduction
• ANC Formalities and the LMS Algorithm
• Applications of ANC
• Summary of ANC Methods

Systems Based on Fundamental Frequency Tracking

• Introduction
• Single-Channel ANC
• Adaptive Comb Filtering
• Harmonic Selection
• Summary of Systems Based on Fundamental Frequency Tracking

Performance Evaluation

• Introduction
• Enhancement and Perceptual Aspects of Speech
• Speech Enhancement Algorithm Performance

Conclusions

Problems

The INTEL System

Addressing Cross-Talk in Dual-Channel ANC

Speech Quality Assessment

Introduction

• The Need for Quality Assessment
• Quality Versus Intelligibility

Subjective Quality Measures

• Intelligibility Tests
• Quality Tests

Objective Quality Measures

• Articulation Index
• Signal-to-Noise Ratio
• Itakura Measure
• Other Measures Based on LP Analysis
• Weighted-Spectral Slope Measures
• Global Objective Measures
• Example Applications

Objective Versus Subjective Measures

Problems

Recognition

The Speech Recognition Problem

Introduction

• The Dream and the Reality
• Discovering Our Ignorance
• Circumventing Our Ignorance

The "Dimensions of Difficulty"

• Speaker-Dependent Versus Speaker-Independent Recognition
• Vocabulary Size
• Isolated-Word Versus Continuous-Speech Recognition
• Linguistic Constraints
• Acoustic Ambiguity and Confusability
• Environmental Noise

Related Problems and Approaches

• Knowledge Engineering
• Speaker Recognition and Verification

Conclusions

Problems

Dynamic Time Warping

Introduction

Dynamic Programming

Dynamic Time Warping Applied to IWR

• DTW Problem and Its Solution Using DP
• DTW Search Constraints
• Typical DTW Algorithm: Memory and Computational Requirements

DTW Applied to CSR

• Introduction
• Level Building
• The One-Stage Algorithm
• A Grammar-Driven Connected-Word Recognition System
• Pruning and Beam Search
• Summary of Resource Requirements for DTW Algorithms

Training Issues in DTW Algorithms

Conclusions

Problems

The Hidden Markov Model

Introduction

Theoretical Developments

• Generalities
• The Discrete Observation HMM
• The Continuous Observation HMM
• Inclusion of State Duration Probabilities in the Discrete Observation HMM
• Scaling the Forward-Backward Algorithm
• Training with Multiple Observation Sequences
• Alternative Optimization Criteria in the Training of HMMs
• A Distance Measure for HMMs

Practical Issues

• Acoustic Observations
• Model Structure and Size
• Training with Insufficient Data
• Acoustic Units Modeled by HMMs

First View of Recognition Systems Based on HMMs

• Introduction
• IWR Without Syntax
• CSR by the Connected-Word Strategy Without Syntax
• Preliminary Comments on Language Modeling Using HMMs

Problems

Language Modeling

Introduction

Formal Tools for Linguistic Processing

• Formal Languages
• Perplexity of a Language
• Bottom-Up Versus Top-Down Parsing

HMMs, Finite-State Automata, and Regular Grammars

A "Bottom-Up" Parsing Example

Principles of "Top-Down" Recognizers

• Focus on the Linguistic Decoder
• Focus on the Acoustic Decoder
• Adding Levels to the Linguistic Decoder
• Training the Continuous-Speech Recognizer

Other Language Models

• N-Gram Statistical Models
• Other Formal Grammars

IWR As "CSR"

Standard Databases for Speech-Recognition Research

A Survey of Language-Model-Based Systems

Conclusions

Problems

The Artificial Neural Network

Introduction

The Artificial Neuron

Network Principles and Paradigms

• Introduction
• Layered Networks: Formalities and Definitions
• The Multilayer Perceptron
• Learning Vector Quantizer

Applications of ANNs in Speech Recognition

• Presegmented Speech Material
• Recognizing Dynamic Speech
• ANNs and Conventional Approaches
• Language Modeling Using ANNs
• Integration of ANNs into the Survey Systems of Section 13.9

Conclusions

Problems

Index

Hardcover; 908 pages