Abstract
This paper addresses the underdetermined blind source separation problem, using a filtering approach. We have developed an extension of the FastICA algorithm which exploits the disparity in the kurtoses of the underlying sources to estimate the mixing matrix and thereafter the recovery of the sources is achieved by employing the ℓ1-norm algorithm. Also, we demonstrate how promising FastICA can be to extract the sources, without utilizing the ℓ1-norm algorithm. Furthermore, we illustrate how this scenario is particularly suitable to the separation of the temporomandibular joint (TMJ) sounds, crucial in the diagnosis of temporomandibular disorders (TMDs). © 2006 British Crown Copyright.