Skip to main content
EURASIP Journal on Advances in Signal Processing

Table 2 Instrumental measures for baseline and proposed approaches averaged over seen noise types of the test set data. Note that the SNR conditions of −5 and 15 dB are unseen during training, whereas the remaining SNRs have been seen. Best two approaches are in boldface, high-complexity reference LSTM-cMSA+CLED-cSA-du excluded

From: Speech enhancement by LSTM-based noise suppression followed by CNN-based speech restoration

SNR

Method

PESQ

STOI

SNRI

-5

Noisy

1.35

0.53

0.00

 

MMSE-LSA

1.38

0.49

3.34

 

SG-jMAP

1.37

0.49

3.92

 

LSTM-IRM

1.60

0.64

11.68

 

LSTM-MSA

1.64

0.65

12.51

 

LSTM-cMSA

1.58

0.64

15.56

 

CED-cSA-du

1.52

0.64

11.45

 

LSTM-cMSA + DNN-cSA

1.59

0.67

16.75

 

LSTM-cMSA + CED-cSA-du

1.61

0.69

16.85

 

LSTM-cMSA + CED-cSA-tr

1.61

0.69

16.99

0

Noisy

1.52

0.65

0.00

 

MMSE-LSA

1.63

0.62

4.18

 

SG-jMAP

1.62

0.62

4.96

 

LSTM-IRM

1.98

0.78

17.02

 

LSTM-MSA

2.07

0.80

19.16

 

LSTM-cMSA

2.02

0.79

24.44

 

CED-cSA-du

1.92

0.80

18.74

 

LSTM-cMSA + DNN-cSA

2.05

0.83

26.38

 

LSTM-cMSA + CED-cSA-du

2.11

0.84

25.92

 

LSTM-cMSA + CED-cSA-tr

2.11

0.84

26.32

5

Noisy

1.77

0.76

0.00

 

MMSE-LSA

1.96

0.74

4.90

 

SG-jMAP

1.97

0.75

5.91

 

LSTM-IRM

2.41

0.87

18.24

 

LSTM-MSA

2.52

0.88

20.11

 

LSTM-cMSA

2.54

0.88

26.04

 

CED-cSA-du

2.42

0.90

21.65

 

LSTM-cMSA + DNN-cSA

2.60

0.91

28.50

 

LSTM-cMSA + CED-cSA-du

2.68

0.92

27.65

 

LSTM-cMSA + CED-cSA-tr

2.69

0.92

28.34

10

Noisy

2.11

0.86

0.00

 

MMSE-LSA

2.35

0.84

5.35

 

SG-jMAP

2.39

0.85

6.66

 

LSTM-IRM

2.83

0.92

17.83

 

LSTM-MSA

2.93

0.93

19.15

 

LSTM-cMSA

3.00

0.93

25.52

 

CED-cSA-du

2.96

0.94

23.91

 

LSTM-cMSA + DNN-cSA

3.07

0.94

28.31

 

LSTM-cMSA + CED-cSA-du

3.17

0.95

27.15

 

LSTM-cMSA + CED-cSA-tr

3.18

0.95

28.14

15

Noisy

2.53

0.93

0.00

 

MMSE-LSA

2.75

0.90

5.38

 

SG-jMAP

2.85

0.91

7.04

 

LSTM-IRM

3.24

0.96

16.63

 

LSTM-MSA

3.32

0.96

18.17

 

LSTM-cMSA

3.39

0.96

23.50

 

CED-cSA-du

3.41

0.97

25.57

 

LSTM-cMSA + DNN-cSA

3.43

0.96

26.42

 

LSTM-cMSA + CED-cSA-du

3.51

0.97

25.03

 

LSTM-cMSA + CED-cSA-tr

3.54

0.97

26.30

Mean

Noisy

1.86

0.75

0.00

 

MMSE-LSA

2.01

0.72

4.63

 

SG-jMAP

2.04

0.72

5.70

 

LSTM-IRM

2.41

0.83

16.28

 

LSTM-MSA

2.50

0.84

17.82

 

LSTM-cMSA

2.51

0.84

23.01

 

CED-cSA-du

2.44

0.85

20.26

 

LSTM-cMSA + DNN-cSA

2.55

0.86

25.27

 

LSTM-cMSA + CED-cSA-du

2.62

0.87

24.52

 

LSTM-cMSA + CED-cSA-tr

2.62

0.87

25.22

 

LSTM-cMSA + CLED-cSA-du [39]

2.66

0.88

25.13

Back to article page