Nonnegative matrix factorization with disjointness constraints for single channel speech separation

This paper addresses the problem of single channel speech separation using nonnegative matrix factorization (NMF) technique. In general, the standard NMF algorithm by itself does not guarantee statistical relationship between the matrices it computes. This leads to poor separation performance. To solve this problem, we propose to enforce disjointness constraint on the standard NMF algorithm in the separation process. The multiplicative update rules of the proposed algorithm are also derived in this paper. The performance of the proposed method is compared with standard NMF algorithm, which is based on the same linear model. The experimental results show that the proposed method achieves a better separation quality than the standard NMF.