Blind Source Separation for Convolutive Mixtures of Speech using Subband processing


NAIST \| Library	Advanced Search	Japanese \| English

Please use this identifier to cite or link to this item: http://hdl.handle.net/10061/8255

Title:	Blind Source Separation for Convolutive Mixtures of Speech using Subband processing
Authors:	Shoko Araki Shoji Makino Robert Aichner Tsuyoki Nishikawa Hiroshi Saruwatari
Issue Date:	Sep-2002
Start page:	195
End page:	202
Abstract:	Subband processing is applied to blind source separation ( BSS) for convolutive mixtures of speech in order to overcome the drawback of frequency-domain BSS. In frequency-domain BSS, we cannot use a long frame size to cover long reverberation for several seconds of speech. This is because we cannot correctly estimate the statistics in each frequency bin if we use a long frame with short observed speech signals. In subband based BSS, (1) we can maintain the number of samples needed to estimate the statistics in each subband by using a moderate number of s山bands, and (2) we can cover long reverberation by using FIR filters in each subband. In the proposed subband BSS, the permutation problem can be solved more easily than in the frequency-domain BSS. Moreover, we can avoid the whitening effect of separated signals which occurs in time-domain BSS.
Description:	SMMSP2002: the 2nd International Workshop on Spectral Methods and Multirate Signal Processing, September 2002.
URI:	http://hdl.handle.net/10061/8255
Text Version:	Publisher
Appears in Collections:	情報科学研究科 / Graduate School of Information Science

Files in This Item:

File	Size	Format
SMMSP_2002_195.pdf	964.23 kB	Adobe PDF	View/Open