Skip to main content
EURASIP Journal on Audio, Speech, and Music Processing

Table 6 WERs (%) of ESniff MMSE

From: Compensation of SNR and noise type mismatch using an environmental sniffing based speech recognition solution

Framework

WER (%)

 

Set A

Set B

Set C

Average

MTR

12.23

13.75

16.42

13.68

ESniff MMSR (Conventional)

8.92

16.64

15.09

13.24

ESniff SNR-MMSR (SNR Mapping)

8.09

15.64

14.55

12.40

ESniff MMSE (M = 128)

9.41

15.58

13.11

12.61

ESniff MMSE (M = 32)

8.95

14.99

12.12

12.00

ESniff MMSE (M = 16)

8.81

14.48

13.24

11.96

ESniff MMSE (M = 8)

8.41

13.60

12.30

11.26

ESniff MMSE (M = 4)

8.09

13.11

11.44

10.76

ESniff MMSE (M = 2)

8.25

13.34

12.47

11.13

  1. The number of Gaussian distributions varies with MMSE performed along with a comparison to MTR, conventional ESniff MMSR and ESniff SNR-MMSR when using the FE feature vectors.
Back to article page