Underdetermined Speech Source Separation Based on Hybrid Clustering

Abstract

Since the number of observed signals is less than the number of source signals, underdetermined blind source separation (UBSS) is a kind of ill-posed blind source separation (BSS) problem. The well-known "two-step approach" based on sparse component analysis (SCA) is the main method to solve this problem, and mixed matrix estimation is a key step. An UBSS algorithm based on mixed clustering is proposed in this paper. Firstly, an affinity propagation (AP) clustering algorithm based on the density deviation sampling is used to estimate the number of source signals and the initial clustering center. Then K-means clustering algorithm is used to estimate the mixing matrix accurately. Finally, the minimum l <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> norm method is used to separate the source signals. This algorithm can not only reduce the computational complexity, but also solve the problem of unknown number of source signals. Experiments on speech signal separation show that the proposed algorithm can improve the estimation accuracy of mixed matrix and has good separation performance.