Skip to main content

Table 6 Summary of speech quality in terms of ΔSNR, LLR, and PESQ for simulated reverberated and noisy speech samples in REVERB-Eval. Mean through all noise types and initial SNR levels conditions evaluated

From: Progressive loss functions for speech enhancement with deep neural networks

 

Reference Systems

 

Progressive Systems

   
 

CNN

ResNet

CNN with WP

CNN with UP

ResNet with WP

ResNet with UP

ΔSNR [dB]

21.29

20.36

10.46

34.36

36.28

22.23

LLR

0.58

0.54

0.83

0.61

0.57

0.58

PESQ

2.24

2.37

1.81

2.14

2.39

2.21

  1. Bold results correspond with the best dataset value, and italic results show the second-best value