Optimal multi-pitch estimation using the EM algorithm for co-channel speech separation

Abstract

The paper addresses the problem of optimally estimating (in the ML sense) the pitch of each of several speakers talking simultaneously. This information is needed in systems which perform co-channel speech separation. We propose a multi-pitch model which is used in conjunction with an EM-based iterative estimation scheme. In addition, the pitch period of each speaker is allowed to vary linearly in the analysis interval, thus offering improved co-channel speech separation. The proposed algorithm is shown to outperform standard pitch detection algorithms, in detecting the pitch of simultaneous speakers.