Skip to main content

Table 5 PESQ 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

PESQ

Noise

Babble

Street

Restaurant

SNR (dB)

− 5

0

5

Avg.

− 5

0

5

Avg.

− 5

0

5

Avg.

Noisy mixture

1.23

1.52

1.83

1.53

1.51

1.83

2.02

1.79

1.66

1.88

2.01

1.85

Bi-LSTM [31]

1.85

1.97

2.44

2.09

1.84

2.02

2.49

2.12

1.98

2.11

2.66

2.25

Bi-CRN [34]

1.92

2.13

2.53

2.19

1.93

2.21

2.57

2.23

2.03

2.21

2.77

2.34

SEGAN [40]

1.99

2.21

2.66

2.29

2.05

2.29

2.68

2.34

2.15

2.38

2.83

2.45

GRN [30]

2.08

2.29

2.71

2.36

2.12

2.45

2.75

2.44

2.23

2.49

2.96

2.56

DCN [38]

2.17

2.38

2.85

2.47

2.22

2.49

2.87

2.52

2.31

2.63

3.04

2.66

DCCRN [35]

2.24

2.51

2.94

2.56

2.37

2.65

2.95

2.66

2.47

2.74

3.11

2.77

TSTNN [41]

2.36

2.62

3.07

2.68

2.48

2.73

3.09

2.76

2.55

2.99

3.25

2.93

MASENet [46]

2.45

2.76

3.13

2.78

2.59

2.83

3.16

2.87

2.68

3.08

3.37

3.04

SADNUNet [47]

2.58

2.83

3.24

2.88

2.66

2.94

3.27

2.96

2.72

3.16

3.46

3.11

MCGN [42]

2.64

2.90

3.32

2.95

2.79

3.11

3.35

3.08

2.81

3.27

3.53

3.20

DBT-Net [51]

2.69

2.97

3.38

3.01

2.84

3.16

3.40

3.13

2.87

3.34

3.59

3.23

TANSCUNet

2.95

3.12

3.52

3.20

2.97

3.31

3.56

3.44

2.98

3.49

3.84

3.44