A new clustering approach for solving the permutation problem in convolutive blind source separation

In this paper we propose a new clustering approach for solving the permutation ambiguity in convolutive blind source separation. After the transformation to the time-frequency domain, the problem of separation of sources can be reduced to multiple instantaneous problems, which may be solved using independent component analysis. The drawbacks of this approach are the inherent permutation and scaling ambiguities, which have to be corrected before the transformation to the time domain. Here, we propose a new method that allows for aligning up to several hundreds of consecutive bins into clusters. The depermutation of these clusters using some known techniques is then much easier than the original problem. The performance of the proposed method is evaluated on real-room recordings.