Skip to main content

Table 3 ASR WERs (%) of our proposed front-end technologies with state-of-the-art back-end strategies for Eval., which can be compared to the original REVERB challenge results. An extended training data (ext.) can be additionally applied to DNN-HMM

From: Front-end technologies for robust ASR in reverberant environments—spectral enhancement-based dereverberation and auditory modulation filterbank features

REVERB challenge Eval.

1ch ASR task

SimData

RealData

  

Utterance-based batch processing mode

Room1

Room2

Room3

Avg.

Room1

Avg.

  

with multi-condition training

Near

Far

Near

Far

Near

Far

 

Near

Far

 

Results from challenge

REVERB baseline

20.84

21.72

23.43

38.59

28.43

44.79

29.62

59.09

55.81

57.45

  

Our submitted results from [12]

13.64

14.93

16.11

25.65

19.77

30.51

20.09

41.97

42.27

42.12

  

Results from [11]

5.90

6.60

7.90

12.20

8.70

13.20

9.10

32.60

32.30

32.50

  

Results from [11] (ext.)

5.10

5.60

6.70

11.50

7.60

11.60

8.00

27.10

27.90

27.50

This study

SGMM

BN{MFCC(L=R=4,#117)}(#42)

6.52

7.30

8.31

13.41

9.86

16.85

10.37

25.07

24.48

24.77

  

BN{MFCC(L=R=4,#117)}(#42)+SE

6.20

7.52

8.70

14.37

9.72

17.14

10.60

24.75

24.10

24.42

  

BN{AMFB(#117)}(#42)

5.78

6.61

8.37

12.60

9.25

14.48

9.51

24.50

24.21

24.35

  

BN{AMFB(#117)}(#42)+SE

6.34

7.32

9.08

13.40

9.67

14.94

10.12

24.38

23.90

24.14

 

DNN

FBANK(L=R=5,#440)

5.66

7.47

8.44

14.11

9.84

17.20

10.45

28.53

28.46

28.49

  

FBANK(L=R=5,#440)+sMBR

5.86

6.91

7.17

11.51

8.70

14.48

9.10

23.22

24.61

23.91

  

FBANK(L=R=5,#440)+sMBR+SE

6.32

6.90

7.83

12.46

8.53

14.51

9.42

22.69

24.47

23.58

  

AMFB-FBANK(L=R=1,#1080)

5.54

6.22

7.57

11.33

8.17

12.98

8.63

24.15

27.99

26.07

  

AMFB-FBANK(L=R=1,#1080)+sMBR

5.71

6.39

7.17

10.53

7.36

11.82

8.16

22.68

23.43

23.05

  

AMFB-FBANK(L=R=1,#1080)+sMBR+SE

5.86

6.42

7.22

10.71

7.40

11.94

8.25

21.93

23.16

22.71

 

DNN+(ext.)

FBANK(L=R=5,#440)

4.90

6.23

6.49

12.88

8.09

16.02

9.09

26.38

26.43

26.40

  

FBANK(L=R=5,#440)+SE(test)

5.22

6.39

7.06

11.65

7.51

13.34

8.52

24.08

25.42

24.75

  

FBANK(L=R=5,#440)+sMBR

4.76

6.00

5.85

11.40

7.61

13.97

8.26

25.36

24.85

25.10

  

FBANK(L=R=5,#440)+sMBR+SE(test)

4.91

6.05

6.41

10.90

7.01

12.83

8.01

25.14

24.78

24.96

  

AMFB-FBANK(L=R=1,#1080)

5.15

5.79

7.28

10.87

8.10

13.25

8.40

23.70

25.05

24.37

  

AMFB-FBANK(L=R=1,#1080)+SE(test)

5.17

5.95

7.14

10.75

8.12

13.19

8.38

23.16

24.38

23.77

  

AMFB-FBANK(L=R=1,#1080)+sMBR

4.93

6.01

6.37

10.26

7.80

12.13

7.91

24.15

24.92

24.53

  

AMFB-FBANK(L=R=1,#1080)+sMBR+SE(test)

5.07

5.86

6.30

10.34

7.66

11.99

7.86

23.89

23.50

23.69