Due to the imprecision for solving the permutation ambiguity problem

the frequency domain methods for convolutive blind separation have their fundamental limitations in separation quality.Without additional priori information

a post-processing approach is proposed to further attenuate spatial interference from competing sources and noise.Firstly

an

M×N

multiple-input multiple-output (MIMO) system is converted into

N

single-input multiple-output (SIMO) systems by splitting spectrogram technique

and then the transfer functions ratios and generalized sidelobe canceller are employed to reconstruct source signals from the

N

SIMO systems respectively.The performance of the proposed algorithm is demonstrated by computer simutation results.