Topics in Acoustic Echo and Noise Control (eBook)

Preface 6
Contents 8
List of Contributors 13
Abbreviations and Acronyms 15
Part I Introduction 20
1 Acoustic Echo and Noise Control – Where did we come from and where are we going? 21
1.1 The Journey to Maturity 21
1.1.1 The Problems to be Solved 21
1.1.1.1 Voice Controlled Switch 22
1.1.1.2 Center Clipper 23
1.1.1.3 Frequency Shift 24
1.1.1.4 Echo Cancellation and Echo Suppression 24
1.1.1.5 Echo Cancellation 26
1.1.1.6 Control of the Filter Adaptation 27
1.1.1.7 Echo and Noise Suppression 28
1.2 State of the Art 28
1.3 Outline of this Book 30
References 33
Part II Multi-Microphone Processing 35
2 Joint Optimization of Acoustic Echo Cancellation and Adaptive Beamforming 36
2.1 Introduction 36
2.2 Concepts for Joint Acoustic Echo Cancellation and Adaptive Beamforming 38
2.2.1 Acoustic Echo Cancellation 38
2.2.2 Adaptive Beamforming 40
2.2.3 Joint Acoustic Echo Cancellation and Adaptive Beamforming 42
2.3 Joint Optimization of Acoustic Echo Cancellation and Adaptive Beamforming 44
2.4 Implementation 51
2.4.1 Robust Generalized Sidelobe Canceller (RGSC) 51
2.4.1.1 Quiescent weight vector 52
2.4.1.2 Blocking Matrix 52
2.4.1.3 Interference Canceller 52
2.4.1.4 Adaptation Control 53
2.4.2 Acoustic Echo Canceller 53
2.4.3 Computational Complexity 53
2.5 Experimental Results 55
2.5.1 Time-Invariant Echo Paths and Time-Invariant Source Position 55
2.5.2 Time-Varying Echo Path and Time-Varying Source Position 57
2.5.3 Reverberation Time 59
2.6 Conclusion 60
References 61
3 Blind Source Separation of Convolutive Mixtures of Audio Signals in Frequency Domain 68
3.1 Introduction 68
3.2 Blind Source Separation for Convolutive Mixtures 70
3.3 Overview of Frequency-Domain Approach 72
3.4 Complex-Valued Independent Component Analysis 75
3.5 Separation Mechanism of Blind Source Separation 77
3.6 Source Localization 78
3.6.1 Basic Theory of Near.eld Model 79
3.6.2 Direction of Arrival Estimation with Far-.eld Model 81
3.7 Permutation Alignment 84
3.7.1 Localization Approach 85
3.7.2 Correlation Approach 86
3.7.3 Integrated Method 87
3.8 Scaling Alignment 89
3.9 Spectral Smoothing 89
3.9.1 Windowing 90
3.9.2 Minimizing Error by Adjusting Scaling Ambiguity 91
3.10 Experimental Results 92
3.10.1 2 × 2, 3 × 3, × 93
3.10.2 6 × 8 with Planar Array 97
3.10.3 2 × 2 Moving Sources 100
3.11 Conclusion 102
References 103
4 Localization and Tracking of Acoustical Sources 107
4.1 Introduction 107
4.2 Source Localization Using the Generalized Cross-Correlation Function 110
4.3 Source Localization Based on Interaural Time Differences 113
4.4 Source Localization Using Adaptive Filters 119
4.5 Some Remarks on Algorithm Selection 126
4.6 Frequency-Domain Adaptive Beamformer with Speaker Tracking 127
4.7 Conclusions 136
Acknowledgements 136
References 137
Part III Echo Cancellation 139
5 Adaptive Algorithms for the Identi.cation of Sparse Impulse Responses 140
5.1 Introduction 140
5.2 Notation and De.nitions 141
5.3 Sparseness Measure 143
5.4 The NLMS, PNLMS, and IPNLMS Algorithms 145
5.5 Universal Criterion 147
5.5.1 Linear Update 148
5.5.2 nonlinear Update 149
5.6 Exponentiated Gradient Algorithms 150
5.6.1 The EG Algorithm for Positive Weights 150
5.6.2 The EG± Algorithm for Positive and Negative Weights 151
5.6.3 The Exponentiated RLS (ERLS) Algorithm 154
5.7 The Lambert W Function Based Gradient Algorithm 155
5.8 Some Important Links Among Algorithms 156
5.8.1 Link Between NLMS and EG± Algorithms 156
5.8.2 Link Between IPNLMS and EG± Algorithms 157
5.8.3 Link Between LWG and EG± Algorithms 159
5.9 Simulations 160
5.10 Conclusions 164
References 166
6 Selective-Tap Adaptive Algorithms for Echo Cancellation 169
6.1 Introduction 170
6.2 Sequential and Periodic Tap Selection 171
6.3 MMax Tap Selection 173
6.3.1 The MMax-NLMS Algorithm 174
6.3.2 Dependence of Convergence Rate on MMax Tap Selection 175
6.3.3 The MMax A.ne Projection Algorithm 176
6.3.4 The MMax Recursive Least Squares Algorithm 177
6.3.5 Computational Complexity 179
6.4 Selective Partial Update Tap Selection 180
6.5 Performance Comparison for Single-Channel 182
Selective-Tap algorithms 182
6.6 Convergence Analysis 182
6.6.1 Non-stationary System Model 183
6.6.2 Mean Square Misalignment for NLMS with 185
6.6.3 Mean Square Misalignment for MMax-NLMS with 187
6.6.4 Simulation Results for single channel NLMS and 189
6.7 Sparse Partial Update NLMS 193
6.8 multichannel Selective-Tap Algorithms for Stereophonic Acoustic Echo Cancellation 195
6.8.1 Overview and Rationale 195
6.8.2 Reducing Interchannel Coherence using Tap Selection 196
6.9 Exclusive Maximum Tap Selection 199
6.9.1 Formulation and Realization using Exhaustive Search 199
6.9.2 E.cient Realization 202
6.10 Exclusive Maximum Adaptive Filters 204
6.10.1 XM-NLMS Algorithm 205
6.10.2 XMNL-NLMS Algorithm 205
6.10.3 XMNL-AP Algorithm 206
6.10.4 XMNL-RLS Algorithm 206
6.11 SAEC Simulation Results 206
6.11.1 Experimental Setup 206
6.11.2 NLMS Simulations 206
6.11.3 AP Simulations 207
6.11.4 RLS Simulations 208
6.12 Discussion and Conclusion 208
A Appendices 211
A.1 Algorithm Summary Tables 211
A.2 Fourth-order Factorization for Zero Mean Gaussian Variables 214
References 214
1, 266–270, New Orleans, LA, UAS, 1990. 215
1, 448–451, Istanbul, Turkey, 2000. 215
5, 373–376, Hong Kong, 2003. 217
1, 950–954, 2004. 217
6, 3685–3688, Seattle, Washington, USA, 1998. 217
52(4), 938–949, 2004. 217
7 Nonlinear Acoustic Echo Cancellation 218
7.1 Introduction 218
7.2 Nonlinear Acoustic Echo Paths 220
7.3 Volterra Filters 224
7.3.1 Application to Cascaded Structures 228
7.3.2 Time-domain Adaptive Volterra Filters 230
7.3.3 Multidelay Adaptive Volterra Filters 239
7.3.4 Application to Real Systems 246
7.4 Power Filters 250
7.4.1 Application to Cascaded Structures 252
7.4.2 Adaptive Orthogonalized Power Filters 256
7.4.3 Application to Real Systems 268
7.5 Conclusions 270
References 272
8 Intelligent Control Strategies for Hands-Free Telephones 275
8.1 Introduction 275
8.1.1 State Representation of a Hands-Free Telephone 275
8.1.2 Combination of Control Algorithms 278
8.2 Fuzzy Systems 279
8.2.1 Classic Versus Fuzzy Detector – Example: The Correlation Coefficient 280
8.2.2 Application of Fuzzy Systems in a step-size Control for an Echo Canceller 283
8.3 Learning Vector Quantization 288
8.3.1 Example: Fuzzy LVQ for State Detection in a Hands-free Telephone 290
8.4 Prerequisites for Automatic Optimization of Control Algorithms: Optimum Step Size and Cost Function 294
8.4.1 Optimum Step size for Network Training 294
8.4.2 Cost Function 297
8.5 Radial Basis Function Network for Step-Size Control Network for State Detection 317
8.6 Radial Basis Function 317
8.6.1 State Classi.cation 318
8.6.1.1 Estimation of Probability of States 318
8.6.1.2 Probability of State Characteristics 323
8.6.2 Reliability of Detectors 327
8.6.3 Conclusions 329
References 330
Part IV Noise Reduction 334
9 Noise Reduction 335
9.1 Introduction 335
9.1.1 Noise and Speech 335
9.1.2 Types of Disturbances, Aim of Reduction 336
9.1.3 Noise-Reduction Approaches 336
9.1.4 Wiener Filter and Spectral Subtraction 337
9.2 Optimum-Filter Design in the Time Domain 339
9.2.1 Mean-Square Error Minimization 339
9.2.2 Approximate FIR-Filter Solution 340
9.3 Wiener-Filter Description in the Frequency Domain 342
9.3.1 Optimum Frequency Response 342
9.3.2 Approximate FIR-Filter Solution 342
9.4 Examples and Filtering E.ects 343
9.4.1 “Low-Pass Signal” plus “Band-Stop Noise” 343
9.4.2 Decaying Spectrum plus Wide-Band Noise 346
9.5 Wiener-Filter Realizations 346
9.6 Spectral Subtraction: Principles and Realization 349
9.6.1 De.nition and Variants 349
9.6.2 Relation with Wiener Filtering 350
9.6.3 Realization 352
9.7 Noise Power Density Spectrum Estimation 353
9.7.1 Noise Measurement in Speech Pauses 354
9.7.2 Continuous Noise Measurements 355
9.7.3 Minimum Statistics 355
9.7.4 Improved Instationarity Tracking 357
9.8 Subtraction and Weighting Rules 358
9.8.1 Magnitude and Power Subtraction 358
9.8.2 Musical Noise 359
9.8.3 Noise Floor, Over-Estimation, and Non-Linear Subtraction 359
9.8.4 Approaches Based on Statistical Models of Signal and Noise 360
9.9 Spectral Analysis and Synthesis 361
9.9.1 DFT and IDFT 361
9.9.2 Generalizations 361
9.9.3 Complex-Modulated Filterbank 362
9.9.4 Real-Valued Filterbanks 365
9.9.5 Non-Equispaced Frequency Bands 366
9.9.5.1 Partial Recombination 367
9.9.5.2 Warped PPN-FFT 367
9.9.5.3 Pruned Tree Structure 370
9.9.5.4 Wavelet-Related Analysis-Synthesis Systems 370
9.9.6 Adaptive Bandwidths 374
9.9.6.1 Motivation 374
9.9.6.2 Possibilities 374
9.9.6.3 Efficient Realization 376
9.10 System Configurations, Experiments, and Comparisons 377
9.10.1 Status 377
9.10.2 Examples 378
9.10.2.1 Uniform vs. Non-uniform Bandwidths 378
9.10.2.2 Fixed vs. Adaptive Bandwidths 382
9.10.2.3 Noise Instationarity 385
9.11 Further Problems and Ideas, Concluding Remarks 386
References 389
T-SA-9(5), 504–512, 2001. 393
7, 126–137, 1999. 394
10 Noise Reduction with Kalman-Filters for Hands-Free Car Phones Based on Parametric Spectral Speech and Noise Estimates 395
10.1 Introduction 395
10.2 Speech and Car Noise Analysis 397
10.2.1 Car Noise Analysis 397
10.2.1.1 Engine Noise 399
10.2.1.2 Wind Noise 399
10.2.1.3 Tyre Noise 400
10.2.2 Speech Analysis 401
10.3 Theoretical Basics 403
10.3.1 Kalman Filters for Colored Noise Signals 403
10.3.2 Parametric Spectral Estimation 409
10.4 Application of Kalman Filters for Noise Reduction 414
10.4.1 Subband Processing 415
10.4.2 AR Model Estimation for Speech and Noise 416
10.4.3 Pitch-Adaptive Enhanced Speech Model Estimation 427
10.5 Comparison of the Results with Classical Frequency Domain Noise Reduction Approaches 430
10.6 Conclusions 435
References 436
Part V Selected Applications 438
11 Evaluation of Algorithms for Speech Enhancement 439
11.1 The Focus of this Chapter 439
11.2 Objective Tests for Noise Suppression 440
11.2.1 Measuring the Quality of Noise Suppression Systems 442
11.2.2 Distance Measures 445
11.2.2.1 Cepstral Distance 445
11.2.2.2 Itakura Measure 448
11.2.2.3 Itakura-Saito Measure 450
11.2.3 Noise Characteristics 451
11.2.3.1 Noise Attenuation 451
11.2.3.2 Musical Noise 451
11.2.3.3 Di.erence in Power Level 452
11.2.4 Psycho-Acoustic Methods 453
11.2.5 Coherence Between Instrumental Measures and Listening Tests 454
11.2.5.1 Rank Correlation 454
11.2.5.2 Judging Quotient 455
11.2.5.3 Results 456
11.3 Comparison Mean Opinion Scores (CMOS) 457
11.3.1 Example 460
11.3.1.1 Basics of Bandwidth Extension Algorithms 460
11.3.1.2 Performing the CMOS Test 463
11.3.1.3 Evaluation of the Test 465
11.3.1.4 Remark 468
11.3.2 Statistical Analysis 469
11.3.2.1 Analysis of the Two-Level Test 469
11.3.2.2 Analysis of the Seven-Level Test 474
11.4 Rhyme Tests 476
11.4.1 Performing a Rhyme Test 478
11.4.2 Example 481
11.4.2.1 Basics of In-Car Communication Systems 481
11.4.2.2 Rhyme Test for In-Car Communication Systems 483
11.4.3 Statistical Analysis 487
11.4.3.1 Hypotheses 487
11.4.3.2 Results 488
11.5 Outlook 489
References 490
2(4), 598–614, 1994. 490
57(4/5), 257–267, 1989 (in German). 492
12 An Auditory Scene Analysis Approach to Monaural Speech Segregation 493
12.1 Introduction 493
12.2 Computational Auditory Scene Analysis 496
12.2.1 Computational Goal of CASA 496
12.2.2 Motivation and Overview of the Approach 497
12.3 Peripheral Analysis and Feature Extraction 498
12.3.1 Auditory Periphery 499
12.3.2 Correlogram and Cross-Channel Correlation 499
12.3.3 Onset and O.set 500
12.3.4 Pitch Determination 502
12.4 Auditory Segmentation 505
12.4.1 Segmentation for Voiced Speech 505
12.4.2 Segmentation Based on Onset/O.set Analysis 506
12.5 Voiced Speech Grouping 507
12.6 Unvoiced Speech Grouping 511
12.6.1 Segregation of Stop Consonants 512
12.6.2 Grouping of Fricatives and A.ricates 513
12.7 Concluding Remarks 516
Appendix: Voiced Speech Segregation Algorithm 517
Acknowledgments 520
References 520
2, 749–752, 2003. 522
1, 357–360, Albuquerque, NM, USA, 1990. 523
13 Wave Field Synthesis Techniques for Spatial Sound Reproduction 524
13.1 Introduction 524
13.2 Elements from the Foundations of Acoustics 525
13.2.1 Coordinate Systems 525
13.2.1.1 Two-Dimensional Coordinates 525
13.2.1.2 Three-Dimensional Coordinates 526
13.2.2 The Wave Equation 526
13.2.2.1 Plane Wave Solution 526
13.2.2.2 Green’s Functions 527
13.2.2.3 Relation Between the Green’s Functions for Line and Point Sources for the Free Field Case 530
13.2.3 Kirchhoff-Helmholtz Integral 532
13.2.3.1 Kirchho.-Helmholtz Integral for a General 3D Volume 532
13.2.3.2 Kirchho.-Helmholtz Integral for a Prism 532
13.3 Wave Field Synthesis 534
13.3.1 Introduction 534
13.3.2 Kirchho.-Helmholtz Integral based Sound Reproduction 535
13.3.3 Monopole and Dipole Sources 537
13.3.4 Reduction to Two Spatial Dimensions 539
13.3.5 Spatial Sampling 540
13.3.6 Driving Signals 541
13.3.6.1 Boundary Conditions 541
13.3.6.2 Determination of the Normal Derivative 543
13.3.6.3 Independence of the Driving Signals from the Listener Position 546
13.3.7 Signal Processing Structure 546
13.4 Implementation of a Wave Field Synthesis System 547
13.5 Conclusions 549
References 550
14 Signal Processing for In-Car Communication Systems 553
14.1 Basics 555
14.1.1 Communication without Intercom Systems 555
14.1.2 Communication with Intercom Systems 560
14.1.2.1 Frequency Response of the Closed-Loop System 561
14.1.2.2 Transmission from Speaking to Listening Passenger 563
14.1.2.3 Transmission from Speaking to Speaking Passenger 564
14.2 Signal Processing for Intercom Systems 566
14.2.1 Processing Structures 568
14.2.2 Preprocessing 568
14.2.3 Beamforming 569
14.2.4 Echo Cancellation 571
14.2.4.1 Cancellation of the Output Signal of the Car Radio 572
14.2.4.2 Cancellation of the Output Signal of the Intercom System 575
14.2.5 Feedback Cancellation 576
14.2.6 Feedback Suppression 577
14.2.7 Combining Feedback Cancellation and Feedback Suppression 579
14.2.7.1 Extensions 582
14.2.8 Gain Control 584
14.2.8.1 Basic Control Structure 584
14.2.8.2 Automatic Gain Control 586
14.2.8.3 Speech Activity Controlled Attenuation 587
14.2.8.4 Adjustment of the Playback Volume 587
14.2.9 Loudspeaker Equalization 588
14.2.10 Further Signal Processing Units 590
14.3 Evaluation of Intercom Systems 591
14.3.1 Subjective Methods 592
14.3.2 Rhyme Tests 592
14.3.3 Comparison Mean Opinion Scores 594
14.3.4 Objective Methods 595
14.3.4.1 Improvements for the Listening Passengers 595
14.3.4.2 Distortions for the Speaking Passenger 598
14.4 A Real System 598
14.5 Conclusions and Outlook 602
References 602
15 Applications of Adaptive Signal Processing Methods in High-End Hearing Aids 605
15.1 Introduction 605
15.2 Directional Microphones 606
15.2.1 First-Order Di.erential Arrays 607
15.2.2 Second-Order Di.erential Arrays 610
15.3 Noise Reduction 614
15.3.1 Long-Term Smoothed, Modulation Frequency Based Noise Reduction 615
15.3.1.1 Theoretical Basis 615
15.3.1.2 Computational E.cient Realization 616
15.3.2 Wiener-Filter Based, Short-Term Smoothed Noise Reduction Methods 618
15.3.3 Ephraim-Malah Based, Short-Term Smoothed Noise Reduction Methods 621
15.3.4 Future Trends 622
15.4 Multi-Band Compression 623
15.4.1 State-of-the-Art 623
15.4.2 Future Trends 625
15.5 Feedback Cancellation 626
15.5.1 Feedback Suppression: Dynamic and Selective Attenuation of Feedback Components 629
15.5.2 Feedback Compensation 629
15.6 Classification 633
15.6.1 Basic Structure of Monaural Classi.cation 633
15.6.2 Binaural Classi.cation 637
15.6.3 Future Trends 637
15.7 Summary 638
References 638
Index 643