Speech Recognition Over Digital Channels: Robustness and Standards

Speech Recognition Over Digital Channels......Page 4
Contents......Page 10
Forward......Page 14
Preface......Page 18
1.1 Introduction......Page 20
1.2 RSR over Digital Channels......Page 23
1.3 Organization of the Book......Page 24
2.1 Introduction......Page 26
2.2 Some General Issues......Page 27
2.3.1 Preprocessing of the Speech Signal......Page 29
2.3.2 Linear Prediction Analysis......Page 30
2.3.4 Cepstral Coefficients......Page 32
2.3.5 Other Features......Page 34
2.4 Vector Quantization......Page 35
2.5.1 Pattern Matching......Page 37
2.5.2 The Statistical Approach......Page 38
2.6.1 Markov Processes and Hidden Markov Processes......Page 39
2.6.2 Definition of a Discrete HMM......Page 41
2.6.3 The Three Basic Problems......Page 42
2.6.5 Simplifications to Continuous HMM Modeling......Page 47
2.7 Application of HMMs to Speech Recognition......Page 48
2.8 Model Adaptation......Page 51
2.8.1 Maximum Likelihood Linear Regression (MLLR)......Page 52
2.8.2 Maximum a Posteriori Linear Regression (MAPLR)......Page 54
2.9.1 Exponential Weighting......Page 55
2.9.2 Bayesian Optimal Classification with Uncertain Data......Page 56
3.2 Mobile and Wireless Networks......Page 60
3.2.2 Example of a Mobile Network Architecture: GSM/GPRS......Page 62
3.2.3 Degradation in Wireless Networks......Page 65
3.2.4 Wireless Channel Models for RSR......Page 69
3.3 IP Networks......Page 73
3.3.1 The TCP/IP Protocol Suite......Page 74
3.3.2 Degradation in IP Networks......Page 77
3.3.3 Lossy Packet Channel Models......Page 79
3.3.4 Implementation of RSR Systems over Packet Networks......Page 84
3.4 The Acoustic Environment......Page 85
3.4.1 Additive Noise......Page 86
3.4.2 Channel Distortion......Page 91
3.4.3 A Model of Environment......Page 93
3.4.4 Probability Distributions of Noisy Speech Features......Page 96
4.1 Introduction......Page 104
4.2.1 Waveform Coders......Page 105
4.2.2 Parametric Coders......Page 110
4.2.3 Pitch Estimation......Page 111
4.2.4 Hybrid Coders......Page 112
4.3.1 Effect of Coding......Page 119
4.3.2 Effect of Tandeming......Page 121
4.3.3 Treatment of the “Coding” Degradation......Page 122
4.4 Recognition from Codec Parameters......Page 124
4.4.1 Robustness of Bitstream-based NSR......Page 125
4.4.3 Alternative Parameter Conversion......Page 130
4.5 Distributed Speech Recognition......Page 131
4.5.2 Vector Quantizers and Product Codes......Page 132
4.5.3 Very Low Bitrate PLP-based Compression for DSR......Page 139
4.5.4 Transform Coders......Page 140
4.5.5 Scalability in DSR Systems......Page 145
4.6 Comparison between NSR and DSR......Page 146
5.1 Introduction......Page 150
5.2.1 Error Detection......Page 152
5.2.2 Error Correction and Unequal Error Protection......Page 154
5.2.3 Example of Channel Coding for NSR: GSM-EFR......Page 157
5.2.4 Frame-Level Interleaving......Page 158
5.2.5 Media-specific FEC......Page 159
5.3 Error Concealment (EC)......Page 160
5.3.1 Interpolation......Page 161
5.3.2 Estimation......Page 164
5.3.3 Recognizer-based EC Techniques......Page 170
5.3.4 Error Concealment for NSR......Page 175
6.1 Introduction......Page 178
6.2 Noise Reduction Techniques......Page 179
6.2.1 Wiener Filters......Page 180
6.2.2 Short-time Spectral Attenuation......Page 186
6.3 Voice Activity Detection......Page 193
6.3.1 Full-band-energy-based VAD......Page 194
6.3.2 Statistical VAD......Page 196
6.3.3 Using Long-term Information......Page 197
6.4.1 Cepstral Mean Normalization......Page 201
6.4.2 Frequency Analysis of Time-filtered Features......Page 204
6.4.4 Cepstral Mean and Variance Normalization......Page 206
6.4.5 Nonlinear Feature Normalization......Page 208
7.1 Introduction......Page 216
7.2.1 Two-stage Mel-warped Wiener Filtering......Page 218
7.2.2 Offset Compensation and Waveform Processing......Page 225
7.3.1 Computation of the Basic Features......Page 226
7.3.2 AFE/XAFE Sampling Frequency Extension to 16 kHz......Page 227
7.3.3 Blind Equalization......Page 230
7.3.4 Voice Activity Detection......Page 231
7.3.5 Pitch and Class Estimation......Page 233
7.4.1 Basic Feature Vector Quantization......Page 236
7.4.2 Extension Features Quantization and Encoding......Page 237
7.4.3 Bitstream/Payload Format and Error Protection......Page 238
7.5.1 Bitstream Decoding and Decompression......Page 240
7.5.2 Error Detection and Concealment......Page 241
7.5.4 Pitch Tracking and Smoothing......Page 242
A Alternative Representations of the LPC Coefficients......Page 244
B Basic Digital Modulation Concepts......Page 246
C.1.1 Linear Block Codes......Page 248
C.1.2 Cyclic Codes......Page 250
C.1.3 Convolutional Codes......Page 251
C.2 Interleaving......Page 253
Bibliography......Page 256
List of Acronyms......Page 268
Index......Page 272