Language Identification Using Spectral and Prosodic Features (eBook)

Preface 6
Contents 8
Acronyms 11
1 Introduction 12
1.1 Introduction 12
1.2 Cues for Language Identification 13
1.3 Types of Language Identification Systems 17
1.3.1 Explicit LID Systems 17
1.3.2 Implicit LID Systems 18
1.4 Challenging Issues in Automatic Language Identification 18
1.5 Objective and Scope of the Book 19
1.6 Issues Addressed in the Book 20
1.7 Organization of the Book 21
References 21
2 Literature Review 23
2.1 Introduction 23
2.2 Review of Explicit LID Systems 24
2.3 Review of Implicit LID Systems 27
2.4 Reasons for Attraction Towards Implicit LID Systems 30
2.5 Motivation for the Present Work 31
2.6 Summary and Conclusions 32
References 32
3 Language Identification Using Spectral Features 37
3.1 Introduction 37
3.2 Speech Databases 38
3.2.1 Indian Institute of Technology Kharagpur Multi-lingual Indian Language Speech Corpus (IITKGP-MLILSC) 38
3.2.2 Oregon Graduate Institute Database Multi-language Telephone-Based Speech (OGI-MLTS) 40
3.3 Features Used for Automatic Language Identification 41
3.4 Development of Language Models 42
3.5 LID Performance on Indian Language Database (IITKGP-MLILSC) 43
3.5.1 Speaker Dependent LID System 43
3.5.2 Speaker Independent LID System 44
3.5.3 Speaker Independent LID System with Speaker Specific Language Models 47
3.6 LID System Using Spectral Features from Pitch Synchronous ƒ 52
3.6.1 Epoch Extraction Using Zero Frequency Filter Method 56
3.6.2 Extraction of the Spectral Features from PSA and GCRs 57
3.6.3 Performance Evaluation 59
3.7 Performance of Proposed Spectral Features on OGI-MLTS Database 61
3.8 Summary and Conclusions 62
References 62
4 Language Identification Using Prosodic Features 64
4.1 Introduction 64
4.2 Extraction of CV Units from Continuous Speech 65
4.3 Prosodic Differences Among Languages 71
4.4 Extraction of Intonation, Rhythm and Stress (IRS) Features from Syllable and Word Levels 71
4.4.1 Intonation 72
4.4.2 Rhythm 75
4.4.3 Stress 76
4.5 Performance Evaluation Using Syllable and Word Level Prosodic Features 77
4.6 Extraction of Prosodic Features from Global Level 78
4.6.1 F0 Contour 79
4.6.2 Duration Contour 79
4.6.3 E Contour 79
4.7 Performance Evaluation Using Global Level Prosodic Features 80
4.8 Performance Evaluation Using Prosodic Features on OGI-MLTS Database 80
4.9 LID Using Combination of Features 82
4.9.1 Performance of LID System Using IRS Features from Syllable and Word Levels 84
4.9.2 Performance of LID System Using Prosodic Features from Syllable, Word and Global Level 84
4.9.3 Performance of LID System Using Spectral and Prosodic Features 86
4.10 Summary and Conclusions 89
References 89
5 Summary and Conclusions 91
5.1 Summary of the Book 91
5.2 Major Contributions of the Book 92
5.3 Scope for Future Work 93
References 94
Appendix ALPCC Features 95
Appendix BMFCC Features 97
Appendix CGaussian Mixture Model (GMM) 101