Multiple Sound Source Separation by Using DOA Estimation and ICA

In this paper, a multiple sound source separation method based on directions of arrival (DOA) and independent component analysis (ICA) fusion is proposed. Firstly, based on the characteristics of microphone arrays, the mixed speech signals of multiple sources are collected and the DOA information of each source is estimated according to single-source point (SSP) algorithm. Secondly, the estimated DOA information is fused with a Gaussian operator to form a modified Gaussian operator. Finally, the fused modified Gaussian operator is used as a guide to apply the ICA algorithm to separate the mixed speech and obtain each speech. The experiments and objective evaluation show that the proposed method can optimize the iteration speed and achieve better separation quality compared with some existing blind source separation (BSS) approaches.