An Efficient Algorithm for Underdetermined Blind Source Separation of Audio Mixtures

This paper proposes a method to separate audio signals when number of sources is more than the number of the sensors used and the number of source signals is not known in advance. The proposed method is for underdetermined blind source separation where time frequency masking is used to remove some of the signals from the mixtures till the number of mixtures is equal to the number of sources and then kurtosis maximization method is used to separate the remaining signals. In this paper we separate four signals from two mixtures under instantaneous conditions. The results obtained are better than conventional methods.