高功能聲訊迴音消除技術之研究

Abstract

本論文除了對聲訊迴音消除系統架構做整體的介紹外，並探討數種適應性演算法在聲訊迴音消除之收斂速度、穩定性及運算量複雜度的表現。為了要加速適應性濾波器的收斂速度，我們針對語音特性，採用消除語音自身相關性的處理架構。在處理立體聲方面，我們採用可以降低兩個輸入訊號間的交互關連性的前處理技術，使得濾波器可以收斂至正確的房間響應係數。另一方面，我們將外插技術應用在聲訊迴音消除系統上，可以有效地降低適應性濾波器收斂所需的時間。

In this thesis, the theories of acoustic echo cancellation (AEC) were studied. The performance in terms of convergence rate, stability and computational complexity of several adaptive algorithms applied to AEC were evaluated. In order to reduce the correlation of speech and speed up the convergence rate of adaptive filters, we adopt the decorrelation methods based on speech properties. For stereophonic AEC system, we adopt the pre-processing techniques to degrade the cross-correlation between the two input signals so that the filters can converge to the real impulse response coefficients during the adaptation. On the other hand, we apply the extrapolating technique to the application of AEC system. The extrapolating techniques can reduce the adaptation steps obviously. 1.1 Background 1.2 Organization of this thesis Chapter 2 An overview of acoustic echo cancellation system 2.1 The room impulse response 2.1.1 The measurement of room impulse response 2.1.2 The property of room impulse response 2.2 The structure of AEC 2.2.1 Monophonic AEC 2.2.2 Stereophonic AEC 2.3 The algorithms of adaptive filter 2.3.1 Time domain adaptive filters 2.3.2 Frequency domain and subband based adaptive filters 2.4 Performance assessment Chapter 3 Acoustic echo cancellation with decorrelation methods 3.1 NLMS algorithm 3.2 NLMS algorithm with signal decorrelation 3.2.1 The architecture of decorrelation filtering 3.2.2 Proposed architecture of decorrelation methods 3.2.3 The decorrelation filters 3.3 Decorrelation with fixed first-order filters 3.3.1 The properties of speech signal 3.3.2 Fixed decorrelation filters 3.3.3 Simulation results 3.4 Decorrelation with adaptive LPC filters 3.4.1 Linear prediction coding (LPC) 3.4.2 Adaptive decorrelation filters 3.4.3 Simulation results Chapter 4 Stereophonic acoustic echo cancellation with pre-processing methods 4.1 Unique solution problem of stereophonic AEC system 4.2 Stereophonic AEC by adding independent noise 4.3 Pre-processing by using nonlinear filters 4.4 Pre-processing by using time-varying all-pass filters 4.5 Modified nonlinear pre-processing methods Chapter 5 Fast AEC Based on Extrapolated Adaptation 5.1 Extrapolation methods based on exponential model 5.1.1 ELMS algorithm 1 5.1.2 ELMS algorithm 2 5.1.3 Comments on the two ELMS algorithms 5.2 Extrapolation methods based on linear model 5.2.1 Linear ELMS algorithm 5.2.2 Simulation results 5.3 Linear ELMS in the application of AEC system 5.4 Modified linear extrapolation methods 5.4.1 Partial extrapolation method 5.4.2 Extrapolation method with interruption 5.5 Complexity Chapter 6 Conclusion Bibliography Autobiography