Contemporary Methods for Speech Parameterization (SpringerBriefs in Speech Technology)

__Contemporary Methods for Speech Parameterization__ offers a general view of short-time cepstrum-based speech parameterization and provides a common ground for further in-depth studies on the subject. Specifically, it offers a comprehensive description, comparative analysis, and empirical performance evaluation of eleven contemporary speech parameterization methods, which compute short-time cepstrum-based speech features. Among these are five discrete wavelet packet transform (DWPT)-based, six discrete Fourier transform (DFT)-based speech features and some of their variants which have been used on the speech recognition, speaker recognition, and other related speech processing tasks. The main similarities and differences in their computation are discussed and empirical results from performance evaluation in common experimental conditions are presented. The recognition accuracy obtained on the monophone recognition, continuous speech recognition and speaker recognition tasks is contrasted against the one obtained for the well-known and widely used Mel Frequency Cepstral Coefficients (MFCC). It is shown that many of these methods lead to speech features that do offer competitive performance on a certain speech processing setup when compared to the venerable MFCC. The last does not target the promotion of certain speech features but instead aims to enhance the common understanding about the advantages and disadvantages of the various speech parameterization techniques available today and to provide the basis for selection of an appropriate speech parameterization in each particular case. Contemporary Methods for Speech Parameterization 1 1 Introduction 2 1.1 Speech Front-End 2 1.2 The Origins of Speech Parameterization 3 1.3 The Technical Mel-Scale 5 1.4 The Critical Band Concept 7 1.5 A Brief Note on Short-Time Speech Parameterization 8 2 Speech Pre-processing 11 2.1 Basics of the Speech Pre-processing 11 2.2 Summary of the Speech Pre-processing Steps 17 3 DFT-Based Speech Parameterization 19 3.1 The LFCC 20 3.2 The MFCC-FB20 23 3.3 The HTK MFCC-FB24 29 3.4 The MFCC-FB40 32 3.5 The HFCC-E-FB29 35 3.6 The PLP-FB19 39 3.7 Comparison Among the Various DFT-Based Speech Features 43 4 DWPT-Based Speech Parameterization 46 4.1 The WPF-SBC 47 4.2 The WPF-FD 52 4.3 The WPF-OBJ 56 4.4 The WPF-OVL 62 4.5 The WPF-ACE 66 4.6 Comparison Among the Various DWPT-Based Speech Features 72 5 Evaluation on the Monophone Recognition Task 75 5.1 The Monophone Recognition Task 76 5.2 Experimental Protocol 78 5.3 Comparative Evaluation on the Monophone Recognition Task 79 6 Evaluation on the Speech Recognition Task 87 6.1 The Speech Recognition Task 87 6.2 Experimental Protocol 89 6.3 Comparative Evaluation on the Speech Recognition Task 90 7 Evaluation on the Speaker Verification Task 92 7.1 The Speaker Verification Task 92 7.2 Experimental Protocol 94 7.3 Comparison Among Six MFCC Implementations 95 7.4 Comparison Among Four DWPT-Based Speech Features 97 8 Conclusion and Outlook 102 9 Links to Code and Further Sources of Information 104 Cover 1 Contemporary Methods for Speech Parameterization 2 1 Introduction 3 1.1 Speech Front-End 3 1.2 The Origins of Speech Parameterization 4 1.3 The Technical Mel-Scale 6 1.4 The Critical Band Concept 8 1.5 A Brief Note on Short-Time Speech Parameterization 9 2 Speech Pre-processing 12 2.1 Basics of the Speech Pre-processing 12 2.2 Summary of the Speech Pre-processing Steps 18 3 DFT-Based Speech Parameterization 20 3.1 The LFCC 21 3.2 The MFCC-FB20 24 3.3 The HTK MFCC-FB24 30 3.4 The MFCC-FB40 33 3.5 The HFCC-E-FB29 36 3.6 The PLP-FB19 40 3.7 Comparison Among the Various DFT-Based Speech Features 44 4 DWPT-Based Speech Parameterization 47 4.1 The WPF-SBC 48 4.2 The WPF-FD 53 4.3 The WPF-OBJ 57 4.4 The WPF-OVL 63 4.5 The WPF-ACE 67 4.6 Comparison Among the Various DWPT-Based Speech Features 73 5 Evaluation on the Monophone Recognition Task 76 5.1 The Monophone Recognition Task 77 5.2 Experimental Protocol 79 5.3 Comparative Evaluation on the Monophone Recognition Task 80 6 Evaluation on the Speech Recognition Task 88 6.1 The Speech Recognition Task 88 6.2 Experimental Protocol 90 6.3 Comparative Evaluation on the Speech Recognition Task 91 7 Evaluation on the Speaker Verification Task 93 7.1 The Speaker Verification Task 93 7.2 Experimental Protocol 95 7.3 Comparison Among Six MFCC Implementations 96 7.4 Comparison Among Four DWPT-Based Speech Features 98 8 Conclusion and Outlook 103 9 Links to Code and Further Sources of Information 105 Cover......Page 1 Contemporary Methods for Speech Parameterization......Page 2 1.1 Speech Front-End......Page 3 1.3 The Technical Mel-Scale......Page 5 1.4 The Critical Band Concept......Page 7 1.4 The Critical Band Concept......Page 8 2.1 Basics of the Speech Pre-processing......Page 11 2.2 Summary of the Speech Pre-processing Steps......Page 17 3 DFT-Based Speech Parameterization......Page 19 3 DFT-Based Speech Parameterization......Page 20 3.2 The MFCC-FB20......Page 23 3.3 The HTK MFCC-FB24......Page 29 3.4 The MFCC-FB40......Page 32 3.5 The HFCC-E-FB29......Page 35 3.6 The PLP-FB19......Page 39 3.7 Comparison Among the Various DFT-Based Speech Features......Page 43 4 DWPT-Based Speech Parameterization......Page 46 4 DWPT-Based Speech Parameterization......Page 47 4.2 The WPF-FD......Page 52 4.3 The WPF-OBJ......Page 56 4.4 The WPF-OVL......Page 62 4.5 The WPF-ACE......Page 66 4.6 Comparison Among the Various DWPT-Based Speech Features......Page 72 5 Evaluation on the Monophone Recognition Task......Page 75 5 Evaluation on the Monophone Recognition Task......Page 76 5.2 Experimental Protocol......Page 78 5.2 Experimental Protocol......Page 79 6.1 The Speech Recognition Task......Page 87 6.2 Experimental Protocol......Page 89 6.2 Experimental Protocol......Page 90 7.1 The Speaker Verification Task......Page 92 7.2 Experimental Protocol......Page 94 7.2 Experimental Protocol......Page 95 7.4 Comparison Among Four DWPT-Based Speech Features......Page 97 8 Conclusion and Outlook......Page 102 9 Links to Code and Further Sources of Information......Page 104 1.2 The Origins of Speech Parameterization......Page 4 1.3 The Technical Mel-Scale......Page 6 1.5 A Brief Note on Short-Time Speech Parameterization......Page 9 2.1 Basics of the Speech Pre-processing......Page 12 2.2 Summary of the Speech Pre-processing Steps......Page 18 3.1 The LFCC......Page 21 3.2 The MFCC-FB20......Page 24 3.3 The HTK MFCC-FB24......Page 30 3.4 The MFCC-FB40......Page 33 3.5 The HFCC-E-FB29......Page 36 3.6 The PLP-FB19......Page 40 3.7 Comparison Among the Various DFT-Based Speech Features......Page 44 4.1 The WPF-SBC......Page 48 4.2 The WPF-FD......Page 53 4.3 The WPF-OBJ......Page 57 4.4 The WPF-OVL......Page 63 4.5 The WPF-ACE......Page 67 4.6 Comparison Among the Various DWPT-Based Speech Features......Page 73 5.1 The Monophone Recognition Task......Page 77 5.3 Comparative Evaluation on the Monophone Recognition Task......Page 80 6.1 The Speech Recognition Task......Page 88 6.3 Comparative Evaluation on the Speech Recognition Task......Page 91 7.1 The Speaker Verification Task......Page 93 7.3 Comparison Among Six MFCC Implementations......Page 96 7.4 Comparison Among Four DWPT-Based Speech Features......Page 98 8 Conclusion and Outlook......Page 103 9 Links to Code and Further Sources of Information......Page 105