On calculating the inverse of separation matrix in frequency-domain blind source separation

Hiroshi Sawada, Shoko Araki, Ryo Mukai, Shoji Makino

Research output: Chapter in Book/Report/Conference proceedingConference contribution

3 Citations (Scopus)

Abstract

For blind source separation (BSS) of convolutive mixtures, the frequency-domain approach is efficient and practical, because the convolutive mixtures are modeled with instantaneous mixtures at each frequency bin and simple instantaneous independent component analysis (ICA) can be employed to separate the mixtures. However, the permutation and scaling ambiguities of ICA solutions need to be aligned to obtain proper time-domain separated signals. This paper discusses the idea that calculating the inverses of separation matrices obtained by ICA is very important as regards aligning these ambiguities. This paper also shows the relationship between the ICA-based method and the time-frequency masking method for BSS, which becomes clear by calculating the inverses.

Original languageEnglish
Title of host publicationIndependent Component Analysis and Blind Signal Separation - 6th International Conference, ICA 2006, Proceedings
Pages691-699
Number of pages9
DOIs
Publication statusPublished - 2006
Externally publishedYes
Event6th International Conference on Independent Component Analysis and Blind Signal Separation, ICA 2006 - Charleston, SC, United States
Duration: 2006 Mar 52006 Mar 8

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume3889 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference6th International Conference on Independent Component Analysis and Blind Signal Separation, ICA 2006
CountryUnited States
CityCharleston, SC
Period06/3/506/3/8

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint Dive into the research topics of 'On calculating the inverse of separation matrix in frequency-domain blind source separation'. Together they form a unique fingerprint.

Cite this