According to this complementary behavior we introduce a new robust algorithm for stereo speech separation which considers both additive and convolutive noise signals to model the MV and binaural cues ...