Blind deconvolution using Bayesian methods with application to the dereverberation of speech

A blind deconvolution algorithm is presented to address the problem of the dereverberation of speech. A Bayesian algorithm is developed for estimating the source, and the problem of ill-conditioning due to long tails of an acoustic impulse response (AIR) is avoided by marginalizing out the unknown channel parameters. The initial samples of the MAP estimate are determined using a stochastic MCMC (Markov chain Monte Carlo) technique, and these estimates are then used in a sequential procedure for estimating the remaining signal. A filterbank implementation is used to reduce the large deconvolution problem into several smaller independent problems. Simulation results are presented to demonstrate the performance of the algorithm applied to the dereverberation of speech.

[1]  James P. Reilly,et al.  The complex subband decomposition and its application to the decimation of large adaptive filtering problems , 2002, IEEE Trans. Signal Process..

[2]  Sylvia Richardson,et al.  Markov Chain Monte Carlo in Practice , 1997 .

[3]  Yingbo Hua,et al.  Blind identification and equalization of FIR MIMO channels by BIDS , 2001, 2001 IEEE International Conference on Acoustics, Speech, and Signal Processing. Proceedings (Cat. No.01CH37221).

[4]  Peter J. W. Rayner,et al.  Blind single channel deconvolution using nonstationary signal processing , 2003, IEEE Trans. Speech Audio Process..

[5]  Peter Green,et al.  Markov chain Monte Carlo in Practice , 1996 .

[6]  Shun-ichi Amari,et al.  Kalman filter and state-space approach to blind deconvolution , 2000, Neural Networks for Signal Processing X. Proceedings of the 2000 IEEE Signal Processing Society Workshop (Cat. No.00TH8501).

[7]  W. Gilks.  Markov Chain Monte Carlo , 2005 .

[8]  Rong Chen,et al.  Blind restoration of linearly degraded discrete signals by Gibbs sampling , 1995, IEEE Trans. Signal Process..