Skip to main content

Table 3 SDR values of all baseline models under seen noises. Proposed model represented by bold and italic letters

From: Sub-convolutional U-Net with transformer attention network for end-to-end single-channel speech enhancement

Metric

SDR

Noise

Babble

Street

Restaurant

SNR (dB)

− 5

0

5

Avg.

− 5

0

5

Avg.

− 5

0

5

Avg.

Noisy mixture

2.12

4.01

5.89

4.01

1.82

3.65

5.62

3.70

2.16

4.18

6.13

4.16

Bi-LSTM [31]

4.32

6.65

7.92

6.30

4.12

6.01

7.39

5.84

4.56

6.88

8.23

6.56

Bi-CRN [34]

4.79

6.94

8.37

6.70

4.51

6.43

7.88

6.27

4.87

7.26

8.6

6.91

SEGAN [40]

5.03

7.22

8.72

6.99

4.92

6.89

8.42

6.74

5.12

7.63

8.91

7.22

GRN [30]

5.25

7.59

9.22

7.35

5.23

7.26

8.87

7.12

5.97

7.97

9.44

7.79

DCN [38]

5.85

7.99

9.64

7.83

5.56

7.84

9.18

7.53

6.22

8.23

9.83

8.09

DCCRN [35]

6.13

8.34

9.96

8.14

5.89

8.26

9.53

7.89

6.79

8.59

10.25

8.54

TSTNN [41]

6.57

8.74

10.42

8.58

6.11

8.59

9.85

8.18

7.16

8.92

10.69

8.92

MASENet [46]

6.94

9.12

10.84

8.97

6.61

8.83

10.21

8.55

7.54

9.36

10.95

9.28

SADNUNet [47]

7.32

9.51

11.11

9.31

6.97

9.15

10.60

8.91

7.89

9.74

11.31

9.65

MCGN [42]

7.61

9.87

11.53

9.67

7.36

9.54

10.94

9.28

8.02

10.09

11.75

9.95

DBT-Net [51]

7.92

10.03

11.82

9.92

7.64

9.86

11.10

9.53

8.22

10.24

11.97

10.14

TANSCUNet

8.62

10.69

12.57

10.63

8.37

10.48

11.58

10.14

8.81

10.86

12.71

10.79