Blind Separation of Noisy Mixed Speech Based on Independent Component Analysis and Neural Network

Blind source separation problem has recently drawn a lot of attention in unsupervised neural learning. In the current approaches, the additive noise is typically negligible so that it can be omitted from the consideration. To be applicable in realistic scenarios, blind source separation approaches should deal evenly with the presence of noise. In this contribution, a noisy multiple channels blind source separation algorithm was proposed based on independent component analysis and neural network. At prewhitening, the data have no noise was used to whiten the noisy data, and the wind age wipe off technique was used to correct the infection of noise, a neural network model having denoise capability was adopted to realize the multiple channels blind source separation method for mixing speech corrupter with white noise. The result shows that this method may reduce the affect of noise and improve the signal-noise ratio (SNR) of separation speech, accordingly renew the original speech.