Abstract
This paper proposes an algorithm for the blind dereverberation of speech signals based on multi-channel linear prediction. Traditional dereverberation methods usually perform well when the input signal is white noise. However, when dealing with colored signals generated by an autoregressive process such as speech, the generating autoregressive process is deconvolved causing excessive whitening of the signal. We overcome this whitening problem by estimating the generating autoregressive process based on multichannel linear prediction and applying this estimated process to the whitened signal so that the input signal can be recovered. Simulation results show the good potential of the proposed method.
