Underdetermined convolutive speech separation method based on channel identification and sparse recovery

In this paper, we consider the problem of separating of speech sources from their underdetermined convolutive mixtures with channel identification and recovery. Our proposed algorithm does not require any prior knowledge of the source geometry or the DOA information. The first step of the proposed algorithm is to estimate the convolutive channel from the speech mixtures after a clustering procedure is implemented to select time interval during which only one source signal is effectively present. The second step is to recover the speech signal based on a compressed sensing (CS) concept to use the sparse structure of the speech signals. Numerical experiments including the comparison with other separation approaches for convolutive speech mixtures are provided to show that our algorithm achieves desirable performance improvement.