Presentation 2001/1/18
Optimization on the Number of Subbands in Frequency-Domain Blind Source Separation
Tsuyoki NISHIKAWA, Shoko ARAKI, Shoji MAKINO, Hiroshi SARUWATARI,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) This paper describes an optimization strategy in terms of the number of subbands in frequency-domain blind source separation (BSS). In general, the separation performance of the conventional ICA-based BSS method significantly degrades under reverberant conditions. On the other hand, as for the inverse filter for dereverberation, it is known that the higher performance can be achieved as the number of filter taps (or the number of subbands) increases. Accordingly, first, we carry out the BSS experiments by increasing the number of subbands in ICA to improve the BSS performance. The results of the signal separation experiments reveal that the separation performance degrades when the number of subbands is exceedingly large; e.g., 1024-or 2048-subband are used. In order to show the cause of the degradation, next, we newly define a simple objective measure to quantify an independence, and investigate the relations between the number of subbands and the independence among narrowband sound sources. The results of the measurements clarify that the independence decreases as the number of subbands increases, and we can conclude that the optimal number of subbands exists in BSS based on the frequency-domain ICA.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Blind source separation / Frequency domain / Subband processing / Independent component analysis / Circular convolution
Paper # US2000-88,EA2000-95
Date of Issue

Conference Information
Committee EA
Conference Date 2001/1/18(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Engineering Acoustics (EA)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Optimization on the Number of Subbands in Frequency-Domain Blind Source Separation
Sub Title (in English)
Keyword(1) Blind source separation
Keyword(2) Frequency domain
Keyword(3) Subband processing
Keyword(4) Independent component analysis
Keyword(5) Circular convolution
1st Author's Name Tsuyoki NISHIKAWA
1st Author's Affiliation Graduate School of Information Science, Nara Institute of Science and Technology()
2nd Author's Name Shoko ARAKI
2nd Author's Affiliation NTT Communication Science Laboratories
3rd Author's Name Shoji MAKINO
3rd Author's Affiliation NTT Communication Science Laboratories
4th Author's Name Hiroshi SARUWATARI
4th Author's Affiliation Graduate School of Information Science, Nara Institute of Science and Technology
Date 2001/1/18
Paper # US2000-88,EA2000-95
Volume (vol) vol.100
Number (no) 580
Page pp.pp.-
#Pages 7
Date of Issue