EURASIP Journal on Advances in Signal Processing

Table 2 System performance on clean data (WER (%))

From: Reverberant speech recognition combining deep neural networks and deep autoencoders augmented with a phone-class feature

	ClnData
	Room 1	Room 2	Room 3	Ave.
Baseline (cln, w/o MLLR)	7.79	8.26	7.80	7.96
Baseline (mc, w/o MLLR)	18.60	18.57	17.80	18.33
Baseline (mc, w MLLR)	13.88	14.01	13.44	13.78
DNN-HMM (cln)	4.02	4.47	4.38	4.29
DNN-HMM (mc)	6.25	6.40	6.38	6.35

Back to article page