Blind decomposition of binaural mixed signals using high-convergence algorithm combining SIMO-ICA and DOA estimation

Summary form only given. In this paper, we address a blind decomposition problem of binaural mixed signals, and we propose a novel blind decomposition algorithm combining frequency-domain ICA (FDICA) with projection back (PB) processing (FDICA-PB), the direction of arrival (DOA) estimation, and single-input-multiple-output-model-based independent component analysis (SIMO-ICA). First, the proposed algorithm estimates the DOA of sources using the outputs of FDICA-PB. Next, the method resets the valid initial value of separation filter and re-optimizes the filter using FDICA-PB and SIMO-ICA algorithm. In this procedure, a filter bank of previously measured head related transfer functions for multiple DOA is supplied to generate the valid initial filter. This self-generation of the initial filter provides the high convergence. To evaluate its effectiveness, decomposition experiments are carried out under a reverberant condition. The experimental results reveal that the decomposition performance of the proposed method is superior to those of the conventional methods.