An effective method on blind speech separation in strong noisy environment

Researches in speech separation that have existed mostly concentrate on the noise-free or low noise situations. But speech communication in reality is usually interfered by strong noise. In this paper we present an effective blind speech separation method in strong noisy environment based on discrete wavelet transform (DWT) and independent component analysis (ICA). Firstly, denoising the strong noisy mixtures with DWT is performed, in which step de-noising is done reasonably in order to get little distorted speech mixtures. Then separate these mixtures by ICA to get each independent speech. Finally, further post processing that includes the normalization and post de-noising has been performed to improve the performance of the separated speeches, that obtains the final estimated sources. Simulation results exhibit a high level of separating results.