Skip to main content
EURASIP Journal on Advances in Signal Processing

Table 6 Comparison of autoencoder target options (WER (%))

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

 

SimData

RealData

DNN-HMM (cln) + DAE (40)

10.03

31.25

DNN-HMM (cln) + DAE (120)

9.29

33.61

DNN-HMM (cln) + DAE (1320)

9.28

34.68

DNN-HMM (mc) + DAE (40)

10.62

24.93

DNN-HMM (mc) + DAE (120)

9.70

26.32

DNN-HMM (mc) + DAE (1320)

9.65

28.21

Back to article page