Aiming at the problem that the suppression effect of traditional speech enhancement algorithm in nonstationary noise

especially in case that in environment where noise is speech

inhibition of noise is decreased sharply

an improved multichannel post-filtering speech enhancement algorithm based on transfer functions generalized sidelobe canceller（ TF-GSC） and optimal modified log spectral amplitude estimator（ OM-LSA） is proposed.The algorithm uses the correlation between the TF-GSC output signal and the reference noise to solve the speech existence probability and update the noise power spectrum estimation in post-filtering. The experimental results show that the algorithm can effectively suppress non-stationary noise and improve the robustness of speech enhancement algorithm in speech noise environment.