Skip to main content

Table 6 SDR values of baselines under unseen noise condition. 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

Train

Airport

Exhibition hall

SNR (dB)

− 5

0

5

Avg.

− 5

0

5

Avg.

− 5

0

5

Avg.

Noisy mixture

1.83

3.52

5.53

3.63

2.11

3.83

5.82

3.92

2.66

3.98

5.94

4.19

Bi-LSTM [31]

3.71

5.75

6.41

5.29

3.81

5.93

6.81

5.52

3.93

5.59

6.89

5.47

Bi-CRN [34]

4.05

6.19

6.75

5.66

4.25

6.37

7.08

5.90

4.27

6.94

7.21

6.14

SEGAN [40]

4.45

6.63

7.29

6.12

4.69

6.71

7.69

6.36

4.71

7.21

7.59

6.50

GRN [30]

4.99

6.89

7.72

6.53

5.12

6.97

7.92

6.67

5.12

7.52

7.89

6.84

DCN [38]

5.52

7.13

8.02

6.89

5.69

7.33

8.29

7.10

5.43

7.85

8.34

7.21

DCCRN [35]

5.81

7.39

8.36

7.22

5.91

7.62

8.67

7.41

5.71

8.11

8.59

7.47

TSTNN [41]

6.15

7.58

8.80

7.51

6.23

7.95

8.93

7.70

6.03

8.43

8.84

7.77

MASENet [46]

6.32

7.87

9.23

7.81

6.54

8.24

9.33

8.03

6.35

8.79

9.14

8.09

SADNUNet [47]

6.61

8.24

9.65

8.17

6.86

8.51

9.65

8.34

6.77

9.02

9.77

8.52

MCGN [42]

6.95

8.62

10.01

8.53

7.17

8.47

10.23

8.62

7.19

9.37

10.47

9.01

DBT-Net [51]

7.42

9.27

10.74

9.14

7.53

8.84

10.80

9.05

7.82

10.14

10.90

9.62

TANSCUNet

7.42

9.81

11.21

9.48

7.73

10.13

11.64

10.89

7.89

10.17

11.93

10.00