Target Speech Extraction based on Orthogonalization of Joint Distribution of Observed Signals

Abstract

This paper proposes a BSS (Blind Source Separation) method for acoustic signals under a noisy environment. The proposed method can separate the unknown source signals based on the orthogonalization of a joint distribution of observed mixture signals. For a target speech extraction, we use a difference between the joint distribution of speech signals and that of stationary noises. From the simulation results, it is found that the proposed method can separate the original source signals and extract the target speech signal under a noisy environment.