Abstract
In this paper the objective is to separate nonstationary heart and lung sounds from their convolutive mixtures in time domain. In order to separate the sources an orthogonal source model and a gradient based optimization have been used to best model the mixing system and best estimate the parameters respectively. Having diagonal or quasi diagonal covariance matrices for different source segments and also having independent profiles/envelops for different sources (which implies nonstationarity of the sources) are the requirements for our convolutive method. We applied the method to synthetically mixed real heart and lung sound signals. Compared to the other methods, the results show the high capability of the method for separating nonstationary heart and lung sound signals. © 2012 IEEE.