Subband-Based Blind Separation for Convolutive Mixtures of Speech

Shoko ARAKI  Shoji MAKINO  Robert AICHNER  Tsuyoki NISHIKAWA  Hiroshi SARUWATARI  

Publication
IEICE TRANSACTIONS on Fundamentals of Electronics, Communications and Computer Sciences   Vol.E88-A   No.12   pp.3593-3603
Publication Date: 2005/12/01
Online ISSN: 
DOI: 10.1093/ietfec/e88-a.12.3593
Print ISSN: 0916-8508
Type of Manuscript: PAPER
Category: Engineering Acoustics
Keyword: 
blind source separation,  speech separation,  convolutive mixtures,  subband processing,  frequency dependent processing,  

Full Text: PDF(893.6KB)>>
Buy this Article




Summary: 
We propose utilizing subband-based blind source separation (BSS) for convolutive mixtures of speech. This is motivated by the drawback of frequency-domain BSS, i.e., when a long frame with a fixed long frame-shift is used to cover reverberation, the number of samples in each frequency decreases and the separation performance is degraded. In subband BSS, (1) by using a moderate number of subbands, a sufficient number of samples can be held in each subband, and (2) by using FIR filters in each subband, we can manage long reverberation. We confirm that subband BSS achieves better performance than frequency-domain BSS. Moreover, subband BSS allows us to select a separation method suited to each subband. Using this advantage, we propose efficient separation procedures that consider the frequency characteristics of room reverberation and speech signals (3) by using longer unmixing filters in low frequency bands and (4) by adopting an overlap-blockshift in BSS's batch adaptation in low frequency bands. Consequently, frequency-dependent subband processing is successfully realized with the proposed subband BSS.