شناسایی گفتار خودکار قوی: پلی به کاربردهای عملی

Robust Automatic Speech Recognition: A Bridge to Practical Applications establishes a solid foundation for automatic speech recognition that is robust against acoustic environmental distortion. It provides a thorough overview of classical and modern noise-and reverberation robust techniques that have been developed over the past thirty years, with an emphasis on practical methods that have been proven to be successful and which are likely to be further developed for future applications. The strengths and weaknesses of robustness-enhancing speech recognition techniques are carefully analyzed. The book covers noise-robust techniques designed for acoustic models which are based on both Gaussian mixture models and deep neural networks. In addition, a guide to selecting the best methods for practical applications is provided. The reader will: * Gain a unified, deep and systematic understanding of the state-of-the-art technologies for robust speech recognition * Learn the links and relationship between alternative technologies for robust speech recognition * Be able to use the technology analysis and categorization detailed in the book to guide future technology development * Be able to develop new noise-robust methods in the current era of deep learning for acoustic modeling in speech recognition * The first book that provides a comprehensive review on noise and reverberation robust speech recognition methods in the era of deep neural networks * Connects robust speech recognition techniques to machine learning paradigms with rigorous mathematical treatment * Provides elegant and structural ways to categorize and analyze noise-robust speech recognition techniques * Written by leading researchers who have been actively working on the subject matter in both industrial and academic organizations for many years Robust Automatic Speech Recognition: A Bridge to Practical Applications 2 Copyright 3 About the Authors 4 List of Figures 6 List of Tables 8 Acronyms 9 Notations 13 Introduction 15 Automatic Speech Recognition 15 Robustness to Noisy Environments 16 Existing Surveys in the Area 16 Book Structure Overview 19 References 20 Fundamentals of speech recognition 22 Introduction: Components of Speech Recognition 22 Gaussian Mixture Models 24 Hidden Markov Models and the Variants 26 How to Parameterize an HMM 26 Efficient Likelihood Evaluation for the HMM 27 EM Algorithm to Learn the HMM Parameters 30 How the HMM Represents Temporal Dynamics of Speech 31 GMM-HMMs for Speech Modeling and Recognition 32 Hidden Dynamic Models for Speech Modeling and Recognition 33 Deep Learning and Deep Neural Networks 34 Introduction 34 A Brief Historical Perspective 36 The Basics of Deep Neural Networks 36 Alternative Deep Learning Architectures 40 Deep convolutional neural networks 41 Deep recurrent neural networks 42 Summary 44 References 45 Background of robust speech recognition 54 Standard Evaluation Databases 54 Modeling Distortions of Speech in Acoustic Environments 56 Impact of Acoustic Distortion on Gaussian Modeling 59 Impact of Acoustic Distortion on DNN Modeling 63 A General Framework for Robust Speech Recognition 68 Categorizing Robust ASR Techniques: An Overview 70 Compensation in Feature Domain vs. Model Domain 70 Compensation Using Prior Knowledge about Acoustic Distortion 71 Compensation with Explicit vs. Implicit Distortion Modeling 72 Compensation with Deterministic vs. Uncertainty Processing 72 Compensation with Disjoint vs. Joint Model Training 73 Summary 73 References 74 Processing in the feature and model domains 77 Feature-Space Approaches 78 Noise-Resistant Features 79 Auditory-based features 79 Temporal processing 81 Neural network approaches 83 Feature Moment Normalization 86 Cepstral mean normalization 86 Cepstral mean and variance normalization 87 Histogram equalization 88 Feature Compensation 91 Spectral subtraction 91 Wiener filtering 92 Advanced front-end 94 Model-Space Approaches 97 General Model Adaptation for GMM 97 General Model Adaptation for DNN 100 Low-footprint DNN adaptation 100 Adaptation criteria 102 Robustness via Better Modeling 103 Summary 106 References 110 Compensation with prior knowledge 119 Learning from Stereo Data 120 Empirical Cepstral Compensation 120 SPLICE 121 DNN for Noise Removal Using Stereo Data 124 Learning from Multi-Environment Data 128 Online Model Combination 128 Online model combination for GMM 128 Online model combination for DNN 130 Non-Negative Matrix Factorization 131 Variable-Parameter Modeling 134 Variable-parameter modeling for GMM 135 Variable-component DNN 136 Summary 140 References 143 Explicit distortion modeling 149 Parallel Model Combination 151 Vector Taylor Series 153 VTS Model Adaptation 154 Distortion Estimation in VTS 155 VTS Feature Enhancement 158 Improvements over VTS 162 VTS for the DNN-Based Acoustic Model 164 Sampling-Based Methods 166 Data-Driven PMC 166 Unscented Transform 166 Methods Beyond the Gaussian Assumption 168 Acoustic Factorization 168 Acoustic Factorization Framework 169 Acoustic Factorization for GMM 169 Acoustic Factorization for DNN 172 Summary 174 References 177 Uncertainty processing 183 Model-Domain Uncertainty 184 Feature-Domain Uncertainty 185 Observation Uncertainty 185 Uncertainty propagation through multilayer perceptrons 186 Joint Uncertainty Decoding 188 Front-End JUD 188 Model JUD 190 Missing-Feature Approaches 191 Summary 194 References 195 Joint model training 199 Speaker Adaptive and Source Normalization Training 201 Model Space Noise Adaptive Training 202 Joint Training for DNN 207 Joint Front-End and DNN Model Training 207 Joint Adaptive Training 207 Summary 210 References 212 Reverberant speech recognition 215 Introduction 215 Acoustic Impulse Response 218 A Model of Reverberated Speech in Different Domains 223 The Effect of Reverberation on ASR Performance 225 Linear Filtering Approaches 226 Magnitude or Power Spectrum Enhancement 229 Feature Domain Approaches 230 Reverberation Robust Features 230 Feature Normalization 231 Model-Based Feature Enhancement 231 Data-Driven Enhancement 233 Acoustic Model Domain Approaches 237 The REVERB Challenge 240 To Probe Further 243 Summary 243 References 245 Multi-channel processing 251 Introduction 251 The Acoustic Beamforming Problem 253 Fundamentals of Data-Dependent Beamforming 257 Signal Model and Objective Functions 257 Generalized Sidelobe Canceller 260 Relative Transfer Functions 262 Multi-Channel Speech Recognition 265 ASR on Beamformed Signals 265 Multi-Stream ASR 266 To Probe Further 268 Summary 269 References 269 Summary and future directions 273 Robust Methods in the Era of GMM 274 Robust Methods in the Era of DNN 280 Multi-Channel Input and Robustness to Reverberation 283 Epilogue 284 References 287 Index 293 A 293 B 293 C 293 D 293 E 294 F 294 G 294 H 295 I 295 J 295 K 295 L 295 M 295 N 296 O 296 P 296 R 297 S 297 T 298 U 298 V 298 W 298 Z 298

Robust Automatic Speech Recognition: A Bridge to Practical Applications establishes a solid foundation for automatic speech recognition that is robust against acoustic environmental distortion. It provides a thorough overview of classical and modern noise-and reverberation robust techniques that have been developed over the past thirty years, with an emphasis on practical methods that have been proven to be successful and which are likely to be further developed for future applications. The strengths and weaknesses of robustness-enhancing speech recognition techniques are carefully analyzed. The book covers noise-robust techniques designed for acoustic models which are based on both Gaussian mixture models and deep neural networks. In addition, a guide to selecting the best methods for practical applications is provided. The reader will:

• Gain a unified, deep and systematic understanding of the state-of-the-art technologies for robust speech recognition
• Learn the links and relationship between alternative technologies for robust speech recognition
• Be able to use the technology analysis and categorization detailed in the book to guide future technology development
• Be able to develop new noise-robust methods in the current era of deep learning for acoustic modeling in speech recognition

• The first book that provides a comprehensive review on noise and reverberation robust speech recognition methods in the era of deep neural networks
• Connects robust speech recognition techniques to machine learning paradigms with rigorous mathematical treatment
• Provides elegant and structural ways to categorize and analyze noise-robust speech recognition techniques
• Written by leading researchers who have been actively working on the subject matter in both industrial and academic organizations for many years

Content: Front Matter,Copyright,About the Authors,List of Figures,List of Tables,Acronyms,NotationsEntitled to full textChapter 1 - Introduction, Pages 1-7 Chapter 2 - Fundamentals of speech recognition, Pages 9-40 Chapter 3 - Background of robust speech recognition, Pages 41-63 Chapter 4 - Processing in the feature and model domains, Pages 65-106 Chapter 5 - Compensation with prior knowledge, Pages 107-136 Chapter 6 - Explicit distortion modeling, Pages 137-170 Chapter 7 - Uncertainty processing, Pages 171-186 Chapter 8 - Joint model training, Pages 187-202 Chapter 9 - Reverberant speech recognition, Pages 203-238 Chapter 10 - Multi-channel processing, Pages 239-260 Chapter 11 - Summary and future directions, Pages 261-280 Index, Pages 281-286