Skip to main content
EURASIP Journal on Advances in Signal Processing

Table 3 WER (%) comparisons on RealData for close-talking conditions

From: Joint training of DNNs by incorporating an explicit dereverberation structure for distant speech recognition

System

Clean

C-E

N-E

Avg

C-T

2.92

10.77

17.44

10.38

Multi-1

3.38

11.59

16.78

10.58

Multi-2

3.13

10.97

16.04

10.05

DNN-JT1

3.09

10.96

15.18

9.74

DNN-JT2

3.15

10.97

15.58

9.90

DNN-JT3

3.27

11.24

16.09

10.20

DNN-JT4

3.03

10.93

14.84

9.60

  1. Clean for Hi-Fi environment, C-E for common environment and N-E for noisy environment
Back to article page