Speech waveform recovery from a reverberant speech signal using inverse filtering of the power envelope transfer function

Shigeki Hirobayashi, Hiroaki Nomura, Tsunehiko Koike, Mikio Tohyama

Research output: Contribution to journalArticlepeer-review

3 Scopus citations

Abstract

Research on recovering the sound source waveform from signals observed in a room with reverberation is important for diagnosis of mechanical failure and improvement of clarity in conference calls. In this paper, the possibility is studied for recovery of the original sound waveform from only the reverberation time information without inverse filtering of the transfer function of the reverberant field. The power envelope of the indoor impulse response is approximated with an exponential function and the envelope of the speech signal is restored by power envelope inverse filtering of the reverberant signal. As a result, the modulation index improvement scale using the difference of the modulation indices of the restored signal and the reverberant signal is improved by 60% in the case of a reverberation time of about 1 s and a modulation frequency of 10 Hz in an experiment using the modulated noise signal in a reverberant field. In the reverberant speech restoration experiment, the distortion of the power envelope can be reduced by about 3 dB when the power envelope distortion rates of the restored speech and the reverberant speech with respect to the original sound are compared. By means of bandwidth separation of the reverberant speech signal, the power envelope distortion of each band can be improved.

ASJC Scopus subject areas

  • Electrical and Electronic Engineering

Fingerprint

Dive into the research topics of 'Speech waveform recovery from a reverberant speech signal using inverse filtering of the power envelope transfer function'. Together they form a unique fingerprint.

Cite this