A study on the method of speech dereverberation based on the MTF concept

摘要

This paper proposes a speech dereverberation method without measuring the implulse response of room acoustics. In this model, the envelopes and carriers are decomposed from speech signal using a constant-bandwidths filterbank and then they are processed in each channel, respectively. About processing of the envelope dereverberation, we also use the power envelope inverse filtering method based on the MTF concept. This is improved to determine adequatly reverberant time even if long silence exist in. About processing of the carrier recovery, we use the method for regenerating carrier using the estimated fundamental frequency (FO) from the reverberant speech. In this paper, it is assumed that the FO is known. The dereverberated speech signal is, then, reconstructed using filterbank from the dereverberated envelopes and the regenerated carriers. It was shown that the proposed method can dereverberante the reverberant speech accurately.