Speech Coding Algorithms

WAI C. CHU earned a PhD in Electrical Engineering from the Pennsylvania State University. His industry experience includes positions at Texas Instruments and various startup companies in the multimedia field. Currently at DoCoMo USA Labs (San Jose, California), he is involved with R&D activities in speech/audio coding, digital signal processing, and multimedia applications.

• Preface xiiiAcronyms xixNotation xxiii1 Introduction 11.1 Overview of Speech Coding 21.2 Classification of Speech Coders 81.3 Speech Production and Modeling 111.4 Some Properties of the Human Auditory System 181.5 Speech Coding Standards 221.6 About Algorithms 261.7 Summary and References 312 Signal Processing Techniques 332.1 Pitch Period Estimation 332.2 All-Pole and All-Zero Filters 452.3 Convolution 522.4 Summary and References 57Exercises 573 Stochastic Processes and Models 613.1 Power Spectral Density 623.2 Periodogram 673.3 Autoregressive Model 693.4 Autocorrelation Estimation 733.5 Other Signal Models 853.6 Summary and References 86Exercises 874 Linear Prediction 914.1 The Problem of Linear Prediction 924.2 Linear Prediction Analysis of Nonstationary Signals 964.3 Examples of Linear Prediction Analysis of Speech 1014.4 The Levinson–Durbin Algorithm 1074.5 The Leroux–Gueguen Algorithm 1144.6 Long-Term Linear Prediction 1204.7 Synthesis Filters 1274.8 Practical Implementation 1314.9 Moving Average Prediction 1374.10 Summary and References 138Exercises 1395 Scalar Quantization 1435.1 Introduction 1435.2 Uniform Quantizer 1475.3 Optimal Quantizer 1495.4 Quantizer Design Algorithms 1515.5 Algorithmic Implementation 1555.6 Summary and References 158Exercises 1586 Pulse Code Modulation and Its Variants 1616.1 Uniform Quantization 1616.2 Nonuniform Quantization 1666.3 Differential Pulse Code Modulation 1726.4 Adaptive Schemes 1756.5 Summary and References 180Exercises 1817 Vector Quantization 1847.1 Introduction 1857.2 Optimal Quantizer 1887.3 Quantizer Design Algorithms 1897.4 Multistage VQ 1947.5 Predictive VQ 2167.6 Other Structured Schemes 2197.7 Summary and References 221Exercises 2228 Scalar Quantization of Linear Prediction Coefficient 2278.1 Spectral Distortion 2278.2 Quantization Based on Reflection Coefficient and Log Area Ratio 2328.3 Line Spectral Frequency 2398.4 Quantization Based on Line Spectral Frequency 2528.5 Interpolation of LPC 2568.6 Summary and References 258Exercises 2609 Linear Prediction Coding 2639.1 Speech Production Model 2649.2 Structure of the Algorithm 2689.3 Voicing Detector 2719.4 The FS1015 LPC Coder 2759.5 Limitations of the LPC Model 2779.6 Summary and References 280Exercises 28110 Regular-pulse Excitation Coders 28510.1 Multipulse Excitation Model 28610.2 Regular-Pulse-Excited–Long-Term Prediction 28910.3 Summary and References 295Exercises 29611 Code-excited Linear Prediction 29911.1 The CELP Speech Production Model 30011.2 The Principle of Analysis-by-Synthesis 30111.3 Encoding and Decoding 30211.4 Excitation Codebook Search 30811.5 Postfilter 31711.6 Summary and References 325Exercises 32612 The Federal Standard Version of CELP 33012.1 Improving the Long-Term Predictor 33112.2 The Concept of the Adaptive Codebook 33312.3 Incorporation of the Adaptive Codebook to the CELP Framework 33612.4 Stochastic Codebook Structure 33812.5 Adaptive Codebook Search 34112.6 Stochastic Codebook Search 34412.7 Encoder and Decoder 34612.8 Summary and References 349Exercises 35013 Vector Sum Excited Linear Prediction 35313.1 The Core Encoding Structure 35413.2 Search Strategies for Excitation Codebooks 35613.3 Excitation Codebook Searches 35713.4 Gain Related Procedures 36213.5 Encoder and Decoder 36613.6 Summary and References 368Exercises 36914 Low-delay CELP 37214.1 Strategies to Achieve Low Delay 37314.2 Basic Operational Principles 37514.3 Linear Prediction Analysis 37714.4 Excitation Codebook Search 38014.5 Backward Gain Adaptation 38514.6 Encoder and Decoder 38914.7 Codebook Training 39114.8 Summary and References 393Exercises 39415 Vector Quantization of Linear Prediction Coefficient 39615.1 Correlation Among the LSFs 39615.2 Split VQ 39915.3 Multistage VQ 40315.4 Predictive VQ 40715.5 Summary and References 418Exercises 41916 Algebraic CELP 42316.1 Algebraic Codebook Structure 42416.2 Adaptive Codebook 42516.3 Encoding and Decoding 43316.4 Algebraic Codebook Search 43716.5 Gain Quantization Using Conjugate VQ 44316.6 Other ACELP Standards 44616.7 Summary and References 451Exercises 45117 Mixed Excitation Linear Prediction 45417.1 The MELP Speech Production Model 45517.2 Fourier Magnitudes 45617.3 Shaping Filters 46417.4 Pitch Period and Voicing Strength Estimation 46617.5 Encoder Operations 47417.6 Decoder Operations 47717.7 Summary and References 481Exercises 48218 Source-controlled Variable Bit-rate CELP 48618.1 Adaptive Rate Decision 48718.2 LP Analysis and LSF-Related Operations 49418.3 Decoding and Encoding 49618.4 Summary and References 498Exercises 49919 Speech Quality Assessment 50119.1 The Scope of Quality and Measuring Conditions 50119.2 Objective Quality Measurements for Waveform Coders 50219.3 Subjective Quality Measures 50419.4 Improvements on Objective Quality Measures 505Appendix A Minimum-phase Property of the Forward Prediction-error Filter 507Appendix B Some Properties of Line Spectral Frequency 514Appendix C Research Directions in Speech Coding 518Appendix D Linear Combiner for Pattern Classification 522Appendix E CELP: Optimal Long-term Predictor to Minimize the Weighted Difference 531Appendix F Review of Linear Algebra: Orthogonality, Basis, Linear Independence, and the Gram–schmidt Algorithm 537Bibliography 542Index 553

“…well equipped with exercises and with procedures which are helpful in implementing the coders…” (Zentralblatt Math, Vol.1041, No.16, 2004)