Blind Acoustic Source Separation Combining Time-Delayed Autocorrelation and 4TH-Order Cumulants

A blind acoustic source separation algorithm is presented which combines second order statistics, time-delayed autocorrelation and 4th order cumulants. Because the 4th order cumulants are not sensitive to the Gaussian signal, the combined method is more applicable for lower SNR with white Gaussian noise. In order to improve the performance, the sensor signals are orthogonalized first using 4th order cumulants, then the mixing matrix can be estimated from the preprocessed data using the combined algorithm. Experiment demonstrates that the combined algorithm is more effective than using only second order statistics, especially in hard environment