Second-order statistics method based on total least-squares for blind separation of stationary sound signals

A blind separation method using second-order statistics based on total least-squares is presented for sound signal separation. The method identifies the mixing matrix through jointly approximate diagonalisation of correlation fun ction matrices. The existence of error in mixing matrix and observation is considered and the modified blind separation model is given. The source signals are recovered while the total least-squares error is minimum. The smaller distortions exist in the reconstructed signals. The proposed method doesn't select nonlinear function and possesses fast computational speed. Finally, the results of experiment analysis made in semi-anechoic chamber demonstrate the effectiveness of the presented method.