Infinite sparse factor analysis for blind source separation in reverberant environments

Kohei Nagira, Takuma Otsuka, Hiroshi G. Okuno

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

Abstract

Sound source separation in a real-world indoor environment is an ill-formed problem because sound source mixing is affected by the number of sounds, sound source activities, and reverberation. In addition, blind source separation (BSS) suffers from a permutation ambiguity in a frequency domain processing. Conventional methods have two problems: (1) impractical assumptions that the number of sound sources is given, and (2) permutation resolution as a post processing. This paper presents a non-parametric Bayesian BBS called permutation-free infinite sparse factor analysis (PF-ISFA) that solves the two problems simultaneously. Experimental results show that PF-ISFA outperforms conventional complex ISFA in all measures of BSS-EVAL criteria. In particular, PF-ISFA improves Signal-to-Interference Ratio by 14.45 dB and 5.46 dB under RT 60∈=∈30 ms and RT 60∈=∈460 ms conditions, respectively.

Original languageEnglish
Title of host publicationLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Pages638-647
Number of pages10
Volume7626 LNCS
DOIs
Publication statusPublished - 2012
Externally publishedYes
EventJoint IAPR International Workshops on Structural and Syntactic PatternRecognition, SSPR 2012 and Statistical Techniques in Pattern Recognition,SPR 2012 - Hiroshima
Duration: 2012 Nov 72012 Nov 9

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume7626 LNCS
ISSN (Print)03029743
ISSN (Electronic)16113349

Other

OtherJoint IAPR International Workshops on Structural and Syntactic PatternRecognition, SSPR 2012 and Statistical Techniques in Pattern Recognition,SPR 2012
CityHiroshima
Period12/11/712/11/9

    Fingerprint

Keywords

  • Blind source separation
  • Infinite sparse factor analysis
  • Non-parametric Bayes
  • Reverberant mixtures

ASJC Scopus subject areas

  • Computer Science(all)
  • Theoretical Computer Science

Cite this

Nagira, K., Otsuka, T., & Okuno, H. G. (2012). Infinite sparse factor analysis for blind source separation in reverberant environments. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 7626 LNCS, pp. 638-647). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 7626 LNCS). https://doi.org/10.1007/978-3-642-34166-3_70