Cue-preserving MMSE Filter for Binaural Speech Enhancement.

It is widely appreciated that the acoustic enhancement function of hearing aids, which is here represented by ambient noise reduction, needs to be designed such that the spatial auditory scene is well preserved, e.g., in terms of target speech and residual noise cues. In this way, we can properly stimulate the brain functions for sound segregation, localization, and recognition. This paper, firstly, stresses two particular forms of spatial cues distortions that occur with bilateral (i.e., independently operating) hearing aids at the left and right ear. We secondly introduce a rigorous derivation of a cue-preserving linear minimum mean-square error (MMSE) filter for acoustic noise reduction based on the general concept of the common spectral gain function. The proposed filter is then applied to a wide range of signal-to-noise ratios and compared with other gain functions, including the bilateral type. We find that our cue-preserving MMSE filter not only maintains the spatial auditory scene, but also turns out advantageous in terms of SNR improvement.