Solving Permutation Problem in Blind Source Separation of Speech Signals: A Method Applicable for Collinear Sources

Blind source separation (BSS) in frequency-domain involves a permutation problem: the permutation ambiguity of ICA in each DFT bin should be aligned so that a separated signal in the time-domain contains frequency components of the same source signal. This paper presents a method for solving the permutation problem, which does not require any information regarding the position of the source signals. Hence for solving the permutation problem, unlike the direction of arrival (DOA) approach, this method can be used even when sources and sensors are collinear. The method uses correlation between the signals in each DFT bin, one of which is partially separated by a time domain BSS method and the other is obtained by a frequency domain BSS method, to solve the permutation problem. We also utilize the correlation between the adjacent bands and harmonic components to fix the permutation for frequency bands which could not be fixed by previous methods so as to make the system more robust