Underdetermined Blind Speech Signal Separation Method Based on the Improved Shortest Path Method

摘要

Underdetermined blind speech signal separation is a widespread case of blind source separation. Underdetermined blind signal separation is the case that the number of sensors is less than the number of sources. A two-step method is used to solve the underdetermined blind speech signal separation, but it has some shortcomings that are weak noise defense, bad signal sparsity, et al. This paper focuses on the study of sparse signals. Then we proposed a smoothing method to improve the shortest path method. The improved shortest path method can remove the false peaks potential function and enhance the accuracy of the mixing matrix estimation. At last, we gained the accurate estimation of the sources by the improved shortest path method.