Blind extraction of underdetermined mixtures using TF Mask

For blind source recovery problem for underdetermined and non-completely sparse mixtures, BE-TF-Mask (Blind Extraction Based on Time Frequency Mask) algorithm is proposed in this paper. The algorithm combines blind extraction algorithm for the non-completely sparse case with the DUET approach. It computes firstly the extraction vector of source and the normal vector of source direction, and then rotates the mixtures and the mixing matrix using a linear transform, and then recovers the source using binary time-frequency mask method. Several speech signal experiments demonstrate its performance and practical.