Blind source separation from single channel audio recording using ICA algorithms

FastICA method has been proposed for blind identification and separation characteristics of components, this paper has made a study of this method in order to measure its performance in the task of separating real audio signals that share the same channel simultaneously. We propose an SCICA algorithm based on FastICA, which allows finding the mixing matrix and its inverse. In this way, it is possible to find representative bases, which after a clustering process, are used as impulse response filters to discriminate source signals. Parameters used in the process identifying sources are studied to improve the results.