NAISTAR
Advanced Search
Japanese | English

naistar (NAIST Academic Repository) >
学術リポジトリ naistar / NAIST Academic Repository naistar >
国際会議発表論文 / Proceedings >
情報科学研究科 / Graduate School of Information Science >

Please use this identifier to cite or link to this item: http://hdl.handle.net/10061/8196

Title: Combining Null Beamformer and Fixed-point ICA for the Blind Separation of Convolutive Mixture of Speech
Authors: Rajkishore Prasad
Hiroshi Saruwatari
Kiyohiro Shikano
Issue Date: Feb-2004
Abstract: In this paper we present a new algorithm, under the general framework of the Blind Signal Separation (BSS), based on the combination of the frequency domain fixed-point Independent Component Analysis (ICA) and Null Beamforming (NBF) for the separation of convolutive mixture of speech signal picked-up by a Iinear microphone array. The functioning of the fixed-point ICA algorithms such as by the non-Gaussianization of the mixed signal by the negentropy maximization or by the kurtosis maximization are based on the heuristic that the mixed signal gains gaussianity, in accordance with the Central Limit Theorem (CLT), in the mixing process and an Independent Component (IC) can be extracted by reversing the process. Interestingly, mixed speech signal fails to comply CLT in every frequency bin which in turn hampers the functioning of such algorithms, ultimately leading to poor separation performance. This paper presents a solution to that problem by combining NBF with the fixed-point ICA algorithm. Results show that if the pro-method have better separation performance, combination improves the overall performance. NBF have better performance in the non-reverberant condition and its performance degrades in the reverberant condition. Thus the combination with NBF is effective in mitigating the above mentioned handicap of the fixed-point ICA only in the less reverberant condition.
Description: International Conference on Intelligent Signal Processing and Robotics, February 2004.
URI: http://hdl.handle.net/10061/8196
Text Version: Publisher
Appears in Collections:情報科学研究科 / Graduate School of Information Science

Files in This Item:

File SizeFormat
ISPR_2004.pdf1.06 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

 

Copyright (c) 2007-2012 Nara Institute of Science and Technology All Rights Reserved.
DSpace Software Copyright © 2002-2010  Duraspace - Feedback