Blind Dereverberation Based on Generalized Spectral Subtraction by Multi-channel LMS Algorithm

A blind dereverberation method based on power spectral subtraction (SS) using a multi-channel least mean squares algorithm was previously proposed. The results of isolated word speech recognition experiments showed that this method achieved significant improvement over conventional cepstral mean normalization (CMN). In this paper, we propose a blind dereverberation method based on generalized spectral subtraction (GSS), which has been shown to be effective for noise reduction, instead of power SS. Furthermore, we extend the missing feature theory (MFT), which was initially proposed to enhance the robustness of additive noise, to dereverberation. The reliability of each spectral component is calculated through the signal-to-reverberation ratio obtained from the spectrum of dereverberant speech based on GSS. The proposed dereverberation method based on GSS with MFT is evaluated on a large vocabulary continuous speech recognition task. The dereverberation method based on GSS with MFT and beamforming achieves a relative word error reduction rate of 11.4% and 32.6% compared to the dereverberation method based on power SS with beamforming and the conventional CMN with beamforming, respectively.