Binaural noise reduction via cue-preserving MMSE filter and adaptive-blocking-based noise PSD estimation
An important objective of binaural noise reduction algorithms is the preservation of the binaural cues. In this paper an extension of the Multi-channel Wiener filter with binaural cue preservation (MWF-ITF) is presented, where the average noise ITF preservation term is replaced by an instantaneous noise ITF preservation term. This framework in addition allows to impose perfect ITF preservation, leading to a hard-constraint formulation. Experimental results show that the proposed technique yields a better performance in preserving the binaural cues of both the noise component and the speech component compared to the MWF-ITF, without degrading the output SNR.