Blind source separation in a noisy environment using super-exponential algorithm

"Super-exponential" methods (SEMs) are attractive algorithms for solving blind signal separation problems. Conventional SEMs are so sensitive to Gaussian noise that they cannot work in a noisy environment. To overcome this drawback, we proposed a new SEM , which does not utilise second-order statistics but only higher-order cumulants. Hence, the proposed SEM becomes robust to Gaussian noise (RSEM). In this paper, mixed signals in a noisy environment are separated in the frequency domain using an adaptive version of RSEM (ARSEM). After separation, noise components are reduced with a speech enhancement technique. We show the results of this simulation and experiment, which demonstrates the effectiveness of the proposed method