A Clustering Approach for Solving the Spatial Aliasing Problem in Convolutive Blind Source Separation

In this paper we propose to extend a recently introduced clustering approach for solving the permutation ambiguity in convolutive blind source separation to a case where spatial aliasing occurs. A well known approach for separation of sources is the transformation to the time-frequency domain, where the task can be reduced to multiple instantaneous problems. While these may be easily solved using independent component analysis, this approach has the drawback of the inherent permutation and scaling ambiguities, which have to be corrected before the transformation to the time domain or otherwise the whole process will fail. Here, we extend an existing clustering approach to cope with the case where spatial aliasing occurs. This is achieved by exploiting the direction information of whole clusters instead of single bins. The performance of the proposed method is evaluated on real-room recordings.