Region-growing permutation alignment approach in frequency-domain convolutive blind source separation

Authors

  • Lin Wang Institute for Microstructural Sciences, National Research Council Canada, Canada
  • Heping Ding Institute for Microstructural Sciences, National Research Council Canada, Canada
  • Fuliang Yin School of Electronic and Information Engineering, Dalian University of Technology, China

Keywords:

Alignment, Apartment houses, Bins, Frequency bands, Frequency domain analysis, Separation, Signal analysis, Alignment methods, Convolutive blind source separation, Frequency bins, Frequency dependence, Frequency domains, Region growing, Separation problems, Short time Fourier transforms, Signal power, Small region, Two-component

Abstract

A new alignment method was proposed to solve the problem in frequency-domain blind source separation (BSS). The new alignment method was based on an inter-frequency dependence measure involving the powers of separated signals. Bin-wise permutation alignment was applied across all frequency bins using the correlation of separated signal powers and the full frequency band was partitioned into small regions. These small regions were based on the bin-wise permutation alignment result and region-wise permutation alignment was performed in a region-growing manner. It was demonstrated that the convolutive separation problem was converted to instantaneous separation in each frequency bin through short-time Fourier transform (STFT). It was also observed that the correlation of bin-wise signal power ratio tended to be high when the two components belonged to the same source.

Downloads

Published

2009-09-01

How to Cite

1.
Wang L, Ding H, Yin F. Region-growing permutation alignment approach in frequency-domain convolutive blind source separation. Canadian Acoustics [Internet]. 2009 Sep. 1 [cited 2021 Dec. 4];37(3):82-3. Available from: https://jcaa.caa-aca.ca/index.php/jcaa/article/view/2144

Issue

Section

Proceedings of the Acoustics Week in Canada