Bandwidth Extension of Speech Signals

Cover -- TOC$Contents -- CH$1 Introduction -- 1.1 Scope of the Book -- 1.2 Nature of Speech Signals -- 1.3 Band Limited Speech Signals -- 1.4 Organization of the Book -- CH$2 Speech Generation -- 2.1 Human Speech Generation Process -- 2.2 Model for the Speech Generation Process -- 2.2.1 Excitation Signal -- 2.2.2 Vocal Tract Filter -- 2.3 Basic Approach for Bandwidth Extension -- CH$3 Analysis Techniques for Speech Signals -- 3.1 Linear Predictive Analysis -- 3.1.1 Autocorrelation Method -- 3.1.2 Covariance Method -- 3.2 Parametric Representations of the Spectral Envelope -- 3.2.1 Tube Model -- 3.2.2 AR-Coefficients -- 3.2.3 Cepstral Coefficients -- 3.2.4 MFCCs -- 3.2.5 Line Spectral Frequencies -- 3.3 Scalar Speech Features -- 3.3.1 Zero Crossing Rate -- 3.3.2 Gradient Index -- 3.3.3 Fundamental Frequency -- 3.3.4 Kurtosis -- 3.3.5 Spectral Centroid -- 3.3.6 Energy Based Features -- 3.4 Distance Measures -- 3.4.1 Log Spectral Deviation -- 3.4.2 RMS-Log Spectral Deviation -- 3.4.3 Cepstral Distance -- 3.4.4 Likelihood Ratio Distance -- 3.4.5 Itakura Distance -- 3.4.6 Itakura8211;Saito Distance -- 3.4.7 Other Spectral Distance Measures -- CH$4 Excitation Signal Extension -- 4.1 Estimation of the Narrowband Excitation Signal -- 4.2 Extension Using Non-Linear Characteristics -- 4.2.1 Half-Way Rectification -- 4.2.2 Full-Way Rectification -- 4.2.3 Quadratic Characteristic -- 4.2.4 Cubic Characteristic -- 4.2.5 Tanh Characteristic -- 4.2.6 Benesty Characteristic -- 4.2.7 Adaptive Quadratic Characteristic -- 4.3 Extension Using Spectral Shifting/Modulation Techniques -- 4.3.1 Fixed Spectral Shifting -- 4.3.2 Adaptive Spectral Shifting -- 4.4 Extension Using Function Generators -- 4.4.1 Sine Generators -- 4.4.2 White Noise Generators -- 4.5 Power Adjustment -- 4.6 Discussion -- CH$5 Broadband Spectral Envelope Estimation -- 5.1 Generation and Preparation of a Training Data Set -- 5.2 Estimation of the Narrowband Spectral Envelope -- 5.3 Estimation of the Broadband Spectral Envelope Using Codebooks -- 5.3.1 Codebook Training and Operation Using AR Coefficients -- 5.3.2 Codebook Training and Operation Using Cepstral Coefficients -- 5.3.3 Codebook Training and Operation Using MFCCs -- 5.3.4 Codebook Training and Operation Using LSFs -- 5.4 Estimation of the Broadband Spectral Envelope Using Neural Networks -- 5.4.1 Training and Operation Using Cepstral Coefficients -- 5.5 Estimation of the Broadband Spectral Envelope Using Linear Mapping -- 5.5.1 Training and Operation Using Cepstral Coefficients -- 5.6 Combined Approach Using Codebook Classification and Linear Mapping -- 5.6.1 Training and Operation Using Cepstral Coefficients -- 5.7 Discussion -- CH$6 Quality Evaluation -- 6.1 Evaluation of the Excitation Signal Extension -- 6.1.1 Objective Quality Criteria -- 6.1.2 Subjective Quality Criteria -- 6.1.3 Rank Correlation Between Objective and Subjective Results -- 6.2 Evaluation of the Envelope Estimation -- 6.2.1 Objective Quality Criteria -- 6.2.2 Subjective Quality Criteria -- 6.3 Discussion -- CH$7 Summary -- A Levinson8211;Durbin Recursion -- B LBG-Algorithm -- C Generalized Delta Rule or Standard Backpropagation -- D Stability Check -- E Feature Distributions -- F Evaluation Results -- G Abbreviations and Symbols -- G.1 Abbr