A new underdetermined blind source separation algorithm under the anechoic mixing model

Abstract

This paper includes mixed matrix estimation of underdetermined blind source separation (UBSS) under the anechoic mixing model and introduces single source points to the anechoic mixing model. First, we adopt a clustering algorithm to get the amplitude in the mixing matrix. Then, utilizing the amplitudes, We use a single source point detection method to determine the presence of only one source signal. This algorithm finds that single source points by using the real and the imaginary parts of the TF coefficients of mixed signals. Finally, using the points of single source cluster to obtain the mixing matrix. The results of our experiments indicate it can calculate the mixing matrix accurately under the condition when there are more sources using the proposed algorithm.