A blind separation method of instantaneous speech signal via independent components analysis

Abstract

Human speech recognition shows remarkable robustness in a variety of listening conditions, including competing talkers, environmental sounds, and background noise. This kind of distinguishing ability is a specific sensing mechanism capacity which is owned by of the internal understanding mechanism of the human voice, it is known as the "cocktail party effect". Blind Source Separation (BSS) is a method to estimate the original signal by using mixed signals observed, which is based on independent component analysis (ICA). The basic principle is to find the hidden factors or the method to calculate the independent data. From the perspective of linear transformation and linear space, the source signals are independent and non-Gaussian, they can be seen as linear space-based signals, while the observed signal was a linear combination of source signals. This study attempted to isolate the effects that energetic masking, defined as the loss of detectable target information due to the spectral overlap of the target and masking signals, has on multi talker speech perception. we will describe a signal source separation method of two-channel stereo instantaneous mixed signal using independent component analysis The main purpose of ICA is to determine a non-orthogonal transformation when the source signal and linear transformation are both unknown, and makes the transformed output as much as possible statistically independent of the various signal components, thus to estimate the basic structure of space or the source signal in the mixed signals observed. The algorithm of this paper is mainly aimed at the non-positive definite instantaneous mixed voice signal without noise, the blind source separation in frequency domain.