Directivity-Dependency-Reduced Blind Source Separation Integrating ICA, Beamforming and Binary Masking

Abstract

A real-time two-stage blind source separation (BSS) method for convolutive mixtures of speech is now being studied by the authors, in which a single-input multiple-output (SIMO)-model-based independent component analysis (ICA) and an SIMO-model-based binary masking are combined. However because of using humanoid robot head or directional microphones to observe the sounds and adopting binary masking in the 2nd stage, the previous two-stage BSS method has directivity dependency for sound source layout. In this paper, we introduce an improved BSS method which uses omni-directional microphones and creates flexible directivity internally instead of the fixed directional microphones while keeping the separation performance. The experimental results reveal that the separation performance can be considerably improved by using the proposed method compared with the conventional BSS methods.