Table 6 Average WER[%] investigating the effect of LDA context sizes [left (L) and right (R)] on the REVERB challenge dev set using eight-channel data

4

11.83

12.20

12.10

12.57

35.62

34.31

34.10

36.22

5

12.14

12.32

12.46

12.72

34.71

35.34

34.44

33.31

6

12.57

12.33

12.56

12.87

35.49

35.29

34.19

35.11

7

12.83

12.94

13.43

13.49

35.13

35.90

35.67

36.00