A new solution for the permutation problem in the frequency-domain BSS using power-ratio correlation

Abstract

This paper proposes a new method to solve the permutation problem in the frequency-domain source separation based on Independent Component Analysis (ICA). Traditionally, the correlation coefficient of separated signal envelopes is calculated to determine if the separated signals originate from the same source. Recently, a new measure has been proposed, which represents the dominance of the separated signals in the mixtures based on the power-ratio, instead of separated signal envelopes. In this paper, we use the interfrequency correlation between measures of power-ratio of the separated signals in the mixtures to solve the permutation problem. Experimental results show that our method provides an almost perfect solution to the permutation problem for the case, where two sources are mixed in a room whose reverberation time is 130 ms.