A Modified Clustering Approach for Solving the Permutation Problem in Convolutive Blind Source Separation

In this paper we propose a modification to a new clusteringapproach for solving the permutation ambiguity in convolu-tive blind source separation. After the transformation to thetime-frequency domain, the problem of separation of convolu-tively mixed sources can be reduced to multiple instantaneousproblems, which may be solved using independent componentanalysis. The drawbacks of this approach are the so calledpermutation and scaling problems, which have to be correctedbefore the transformation to the time domain. Here, we usea new method that allows for aligning up to several hundredsof consecutive bins into clusters and propose a modificationwhich allows for an even more effective clustering. Thedepermutation of these clusters using some known techniquesis then much easier than the original problem.