Complex cepstrum based single channel speech dereverberation

Abstract

A new method of blind dereverberation of single channel is proposed to solve the existing problem about the study of speech signal. Firstly, segmentation of speech signal with variant length is proposed on the sampled speech signal to obtain relatively accurate estimation of the subsequent cepstrum, and the complex cepstrum is applied to the valid segments. Then, the pre-estimation of the room impulse response is obtained, and so is the inverse-filter solution. Finally, single channel blind deconvolution is used by taking the inverse-filter as the initial parameter of the algorithm. Simulations and experiments demonstrate its validity and robusty.