Underdetermined Blind Separation for Speech Signal Based on Two-Step Sparse Component Analysis

In order to solve the problem of underdetermined blind separation for speech signal, which the classical algorithm such as ICA can´t solve, a blind separation algorithm based on two-step sparse component analysis is proposed. First, it transforms the mixed-voice signal to frequency domain by STFT for sparse representation; then obtains the cluster centers by fuzzy C-Means algorithm and estimates the mixing matrix; finally, recovers the source signals using the shortest path decomposition algorithm according to the mixing matrix. Using similarity coefficient matrix as the separation effect standard, simulation experiment results show that the two-step sparse component analysis can solve the problem of underdetermined blind separation for speech signal.