Blind Speech Separation in the Frequency Domain Using Assignment Problem Approach

Abstract

In this paper, we propose a method for blind speech separation of convolutive mixtures in the frequency domain. The main difficulty in a frequency approach is the so-called permutation problem. In the proposed method, we use the Assignment Problem (AP) approach to solve permuation ambiguity in the frequency domain. In our work, we apply three different algorithms including Hugarian, Jonker-Volgenant and Bertsekas’s Auction to solve the AP to find the optimal solution. Computer simulation experiments are presented to illustrate the proposed method.