A speech dereverberation method based on the MTF concept in power envelope restoration

Abstract

Abstract:We previously proposed an improved method for restoring the power envelope from a reverberant signal, based on the modulation transfer function (MTF) concept in order to resolve the problems of Hirobayashi's method. In this paper, to apply our improved method to reverberant speech, we consider three issues related to speech applications: (i) how to apply the improved method to speech dereverberation based on co-modulation characteristics; (ii) whether the MTF concept can also be applied in the sub-band for reverberant signals; and (iii) whether power envelope inverse filtering should be done separately in each channel. We propose an extended filterbank model based on these considerations. We have carried out 15,000 simulations of the power envelope restoration for reverberant speech signals, and our results have shown that the proposed model can adequately restore the power envelopes in all channels from reverberant speech signals. We also found that the estimation of the reverberation time should be done separately in each channel to improve the restoration accuracy of the power envelope.

Comparative evaluation of modulation-transfer-function-based blind restoration of sub-band power envelopes of speech as a front-end processor for automatic speech recognition systems

“…In the processing scheme, observed signal yðtÞ (a pre-emphasized signal of the original signal with a coefficient of 0.97) is decomposed into a series of frequency sub-bands; envelope detectors then extract temporal modulation envelopes e 2 y;n ðtÞ. Considering the co-modulation characteristics of speech signals in the sub-bands [22], we deliberately designed a series of FIR-type band-pass filters with a constant bandwidth (100 Hz was chosen in this study) for the decomposition (see Subchapter 4.1 for bandwidth selection). Thus, this filterbank is referred to as a constant-bandwidth filterbank (CBFB) in this paper.…”

“…To obtain sub-band power envelopes of clean speech, only inverse MTF (IMTF) filtering is needed because of the relationships between the power envelopes of sub-band reverberant speech, of sub-band clean speech, and of the impulse response of room acoustics. Therefore, this method of restoration does not need the impulse response of room acoustics to be measured to derive inverse filtering [21,22].…”

“…We previously proposed a sub-band power envelope inverse filtering method based on the MTF concept [21][22][23]. We tested its effectiveness in restoring the temporal power envelopes of reverberant signals using correlation and SNR measurements [21,22].…”

“…We tested its effectiveness in restoring the temporal power envelopes of reverberant signals using correlation and SNR measurements [21,22]. These tests demonstrated that the proposed method improves the accuracy of power-envelope restoration and improves speech intelligibility [23].…”

Blind dereverberation using correlation coefficients considering periodicity of voiced speech

“…On the other hand, temporal envelope inverse filtering methods have been proposed [4]- [6]. These methods are based on the modulation transfer function (MTF) concept [7] and restore the temporal envelope from reverberation signal without the room impulse response.…”