Acoustic signal separation of statistically independent sources using multiple microphones

A method for separating acoustic sources of sound that are statistically independent from multiple linear combinations of filtered versions of the sounds is presented. Whereas the covariance statistic can be used to achieve processed signals that are linearly independent, linear combinations of the sources, higher-order statistics can be used to achieve processed signals that contain only a signal from each source alone. It is shown that the fourth-order joint moment is an appropriate statistic to use as a measure of statistical independence. Results of simulations are presented that illustrate the potential of this approach.<>