Sparse component analysis based on an improved ant K-means clustering algorithm for underdetermined blind source separation

This paper proposed an improved sparse component analysis (SCA) approach to improve the performance of underdetermined blind source separation for the acoustic/speech sources. First, the pre-processing to build a sparse model for acoustic/speech sources is described. Then, the proposed SCA approach designs an improved K-means clustering algorithm based on ant colony algorithm to estimate the mixture matrix, and utilize a method based on orthogonal matching pursuit algorithm to re-cover the signals. Experiment results demonstrate that the improved clustering algorithm can enhance the global searching ability which benefits for the proposed SCA approach to achieve a higher estimation accuracy of mixture matrix, and the improved method in the second step can make the proposed SCA approach work well for recovering the multi-channel blind source signals.