A Curtosis Based Criterion for Solving the Permutation Ambiguity in Convolutive Blind Source Separation

In this work, we present a modification of an algorithm forsolving the permutation ambiguity in convolutive blind sourceseparation. A well known approach for separation of convolu-tive mixtures is the transformation to the time-frequency do-main, where the convolution becomes a multiplication. Withthis approach it is possible to use well known instantaneousICA algorithms independently in each frequency bin. Thissimplification leads to reduced computational costs and betterseparation in each frequency bin. However, this simplificationhas the major drawback of arbitrary permutation in eachfrequency bin. Without a correction of this permutationthe restored time domain signals still remain mixed. Anoften used approach for solving this permutation problem isthe dyadic sorting, where groups of bins are consecutivelydepermuted. By recursively joining growing groups all binsgets sorted. In recent works we presented a criterion forthe depermutation, which was based on sparsity in the timedomain of the restored subband signals. In this work wemodify this approach to use a curtosis based criterion which isan alternative measurement for the non-gaussianity of speechsignals.