a tutorial on hidden markov models and selected applications in speech recognition

33
2004/11/16 1 A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition LAWRENCE R. RABINER, FELLOW, IEEE Presented by: Chi-Chun Hsia

Upload: curt

Post on 25-Jan-2016

64 views

Category:

Documents


1 download

DESCRIPTION

A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition LAWRENCE R. RABINER, FELLOW, IEEE Presented by: Chi-Chun Hsia. Markov Chain. Markov Chain. Markov Chain. It results in a Geometric Distribution. And then, what does “hidden” means?. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 1

A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

LAWRENCE R. RABINER, FELLOW, IEEEPresented by: Chi-Chun Hsia

Page 2: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 2

Markov ChainMarkov Chain

Page 3: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 3

Markov ChainMarkov Chain

Page 4: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 4

Markov ChainMarkov Chain

And then, what does “hidden” means?

It results in a Geometric Distribution

Page 5: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 5

Extension to Hidden Markov ModelExtension to Hidden Markov Model

Page 6: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 6

Extension to Hidden Markov ModelExtension to Hidden Markov Model

Page 7: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 7

Elements of an HMMElements of an HMM

Page 8: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 8

The Three Basic ProblemsThe Three Basic Problems

Page 9: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 9

Solution to Problem 1Solution to Problem 1

Page 10: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 10

Solution to Problem 1Solution to Problem 1

Page 11: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 11

Forward-Backward ProcedureForward-Backward Procedure

Page 12: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 12

Forward-Backward ProcedureForward-Backward Procedure

Page 13: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 13

Forward-Backward ProcedureForward-Backward Procedure

Page 14: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 14

Forward-Backward ProcedureForward-Backward Procedure

Page 15: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 15

Solution to Problem 2Solution to Problem 2

Page 16: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 16

Solution to Problem 2Solution to Problem 2

Page 17: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 17

Viterbi AlgorithmViterbi Algorithm

Page 18: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 18

Viterbi AlgorithmViterbi Algorithm

Page 19: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 19

Solution to Problem 3Solution to Problem 3

Page 20: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 20

Solution to Problem 3Solution to Problem 3

Page 21: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 21

EM Algorithm for HMMEM Algorithm for HMM

X.D. HUANG, Y. ARIKI, M.A. JACK

HIDDEN MARKOV MODELSFOR SPEECH RECOGNITION

EDINBURGH UNIVERSITY PRESS

Page 22: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 22

Types of HMMsTypes of HMMs

Page 23: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 23

Continuous Type HMMsContinuous Type HMMs

Page 24: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 24

Autoregressive HMMsAutoregressive HMMs

Page 25: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 25

Optimization CriterionOptimization CriterionMaximum Likelihood (ML)

Maximum Mutual Information (MMI)

Minimum Discrimination Information (MDI)

Minimum Classification Error (MCE) Chang.

Page 26: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 26

Implementation Issues for HMMsImplementation Issues for HMMs

• Scaling

• Multiple Observation Sequences

• Initial Estimates of HMM Parameters

• Effect of Insufficient Training Data

• Choice of Model

Page 27: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 27

ScalingScaling

Page 28: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 28

ScalingScaling

Page 29: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 29

ScalingScaling

And so on and on and on and on……………..

Page 30: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 30

Multiple Observation SequencesMultiple Observation Sequences

Page 31: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 31

Initial Estimates of HMM ParametersInitial Estimates of HMM Parameters

Page 32: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 32

Effect of Insufficient Training DataEffect of Insufficient Training Data

Page 33: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition

2004/11/16 33

Choice of ModelChoice of Model