Statistical Methods for Speech RecognitionThis book reflects decades of important research on the mathematical foundations of speech recognition. It focuses on underlying statistical techniques such as hidden Markov models, decision trees, the expectation-maximization algorithm, information theoretic goodness criteria, maximum entropy probability estimation, parameter and data clustering, and smoothing of probability distributions. The author's goal is to present these principles clearly in the simplest setting, to show the advantages of self-organization from real data, and to enable the reader to apply the techniques. |
Contents
Chapter | 3 |
Formulation | 4 |
Chapter | 15 |
Transition Sequence | 21 |
Parameters of HMMS | 27 |
Normalization | 33 |
The Acoustic Model | 39 |
Forms | 48 |
Chapter 8 | 137 |
References | 145 |
Acoustic Processor | 152 |
Dimensions | 158 |
Decision Trees and Tree Language 10 1 Introduction | 165 |
Chous Method | 179 |
Based on Word Encoding | 184 |
Data | 190 |
References | 52 |
References | 54 |
Model | 60 |
Language Model | 66 |
Personalization from Text | 73 |
Concept | 76 |
Chapter 5 | 79 |
State Spaces | 86 |
Chapter 6 | 93 |
Search | 99 |
Shortcuts | 109 |
Entropy | 119 |
Theorem | 126 |
Generation from Spelling | 197 |
References | 205 |
Maximum Entropy Probability | 211 |
Model | 227 |
Appropriate Constraints | 233 |
Problems | 240 |
Adaptation to a New Domain | 246 |
Model | 253 |
Estimation of Probabilities from Counts | 257 |
Estimate | 263 |
of GoodTuring Estimation | 269 |
279 | |
Other editions - View all
Common terms and phrases
acoustic model acoustic processor alignment allophones applied arg max basic Baum algorithm Baum-Welch algorithm beam search belonging bigram building blocks chapter classification code word composite HMM compute concatenation conditional entropy Conference on Acoustics constraints context corresponding criterion decision tree decoding defined denotes derived determined domain elementary HMMs encoding equivalence classes estimate evaluate fast match formula function held-out hidden Markov models histories h HMM of figure IEEE Transactions Information Theory iterative L.R. Bahl language model leaf Markov chain maximize maximum entropy method mutual information node null transitions observed output symbol path Po(t possible probability distribution problem question R.L. Mercer recognizer recursion relative frequencies result segments Signal Processing Speaker Recognition specified speech recognition split stack statistical subset tion training data trigram language model triphone Viterbi algorithm w₁ w₂ word sequence word string