Blind acoustic echo cancellation without double-talk detection

The near-end and far-end signals in an echo cancellation problem are modeled as independent Gaussian random processes obeying a structured form of non-stationarity in which the statistics are constant over short intervals but are allowed to change from one interval to the next. With these assumptions the echo cancellation problem is cast as a semi-blind source separation problem and principles of maximum likelihood are applied to estimate the echo cancellation filter. The form of this estimate is analyzed as the width of the intervals decreases to a single sample. This leads to an adaptive filtering algorithm reminiscent of recursive least squares (RLS) but enjoying superior convergence properties due to the presence of a normalizing factor that regulates the flow of information into the statistical quantities used by RLS. The normalization acts intrinsically as a “soft” double-talk detection mechanism. Simulation results demonstrate the behavior of the algorithm during multiple conditions including double talk. It achieves 10-15 dB lower misadjustment than conventional RLS adaptive filtering.

[1]  G.S. Moschytz,et al.  Combined blind/nonblind source separation based on the natural gradient , 2001, IEEE Signal Processing Letters.

[2]  Lang Tong,et al.  Indeterminacy and identifiability of blind identification , 1991 .

[3]  Thomas M. Cover,et al.  Elements of Information Theory , 2005 .

[4]  Walter Kellermann,et al.  TRINICON for Dereverberation of Speech and Audio Signals , 2010, Speech Dereverberation.

[5]  Ted S. Wada,et al.  USE OF DECORRELATION PROCEDURE FOR SOURCE AND ECHO SUPPRESSION , 2008 .

[6]  Walter Kellermann,et al.  A real-time blind source separation scheme and its application to reverberant and noisy acoustic environments , 2006, Signal Process..

[7]  Dinh-Tuan Pham,et al.  Blind separation of instantaneous mixtures of nonstationary sources , 2001, IEEE Trans. Signal Process..

[8]  Philippe Garat,et al.  Blind separation of mixture of independent sources through a quasi-maximum likelihood approach , 1997, IEEE Trans. Signal Process..

[9]  Walter Kellermann,et al.  Speech Dereverberation by Blind Adaptive MIMO Filtering Exploiting Nongaussianity, Nonwhiteness, and Nonstationarity , 2010, Sprachkommunikation.

[10]  Kiyotoshi Matsuoka,et al.  A neural net for blind separation of nonstationary signals , 1995, Neural Networks.

[11]  Jacob H. Gunther,et al.  Learning Echo Paths During Continuous Double-Talk Using Semi-Blind Source Separation , 2012, IEEE Transactions on Audio, Speech, and Language Processing.

[12]  Walter Kellermann,et al.  A generalization of blind source separation algorithms for convolutive mixtures based on second-order statistics , 2005, IEEE Transactions on Speech and Audio Processing.

[13]  Ted S. Wada,et al.  Batch-Online Semi-Blind Source Separation Applied to Multi-Channel Acoustic Echo Cancellation , 2011, IEEE Transactions on Audio, Speech, and Language Processing.