Robust MVDR-based feature extraction for speech recognition

This paper presents a novel noise robust feature extraction method for speech recognition. It is based on making the Minimum Variance Distortionless Response (MVDR) power spectrum estimation method robust against noise. This robustness is obtained by modifying the distortionless constraint of the MVDR spectral estimation method via weighting the subband power spectrum values based on the sub-band signal to noise ratios. The above method, when evaluated on Aurora 2 task, outperformed both the MFCC features as the baseline and the MVDR-based features in different noisy conditions.