Measure of Quality of Source Separation for Sub- and Super-Gaussian Audio Mixtures

Conventional Blind Source Separation (BSS) algorithms separate the sources assuming the number of sources equals to that of observations. BSS algorithms have been developed based on an assumption that all sources have non-Gaussian distributions. Most of the instances, these algorithms separate speech signals with super-Gaussian distributions. However, in real world examples there exist speech signals which are sub-Gaussian. In this paper, a novel method is proposed to measure the separation qualities of both super-Gaussian and sub-Gaussian distributions. This study measures the impact of the Probability Distribution Function (PDF) of the signals on the outcomes of both sub and super-Gaussian distributions. This paper also reports the study of impact of mixing environment on the source separation. Simulation improves the results of the separated sources by 7 dB to 8 dB, and also confirms that the separated sources always have super-Gaussian characteristics irrespective of PDF of the signa ls or mixtures.