Robust Constrained Mfmvdr Filtering for Single-Microphone Speech Enhancement
Dörte Fischer, Simon Doclo · 2018
The multi-frame minimum variance distortionless response (MFMVDR) filter for single-microphone speech enhancement exploits speech correlation across consecutive time frames. This filter is designed to avoid speech distortion while minimizing the total signal output power. The MFMVDR filter is very sensitive to estimation errors in the speech correlation vector, since correlated speech components may be mistakenly suppressed. Inspired by robust beamforming approaches, in this paper we propose a robust constrained MFMVDR filter for single-microphone speech enhancement by estimating the speech correlation vector that maximizes the total signal output power within a spherical uncertainty set. For the upper bound of the spherical uncertainty set, we propose to use a trained mapping function that depends on the a-priori SNR. Experimental results for different noise types and SNRs show that the proposed robust approach achieves a more accurate estimate of the speech correlation vector resulting in low speech and noise distortion but a more conservative noise reduction.