Fig. 8From: Joint training of DNNs by incorporating an explicit dereverberation structure for distant speech recognitionAn example of multi-channel front-end: a static LMFB features of one channel reverberant speech; b static LMFB features of WPE+Beamforming processed speech; c the output of front-end regression DNN: before joint training (estimated clean LMFB features)Back to article page