Blind separation of non-linear convolved speech mixtures

In this paper we present a novel solution to the convolutive and post non-linear Blind Speech Separation (NLBSS) problem based on a neural network topology. The non-linear separating functions are chosen to be a mixture of parametric sigmoid functions. The estimation of the separating filter coefficients and the parameters of the separating non-linear functions is derived using the Maximum Likelihood Estimation principle. Extensive experiments using complex non-linear mixing functions and real room impulse responses were carried out to simulate a mixing scenario of two simultaneous speakers in a real room environment under the effect of high non-linear distortions. The proposed method succeeds in separating the non-linearly mixed signals and improves the phoneme recognition accuracy of an automatic speech recognition system by more than 20% in comparison to the accuracy measured with the non-linear mixture signals. Furthermore, this method outperforms standard linear ESS methods by 20%, justifying the necessity for non-linear separating functions integration in speech recognition systems in a non-linear and multi-simultaneous speaker environment.