Skip to main content

Table 5 Speech quality in terms of SRMR, LLR, and PESQ for simulated and real reverberated speech. The last row represents the mean and standard deviation along the experiments presented in each column

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

   Reference systems     Progressive systems    
Dataset   Unproc. WPE [32] CNN ResNet P-CNN P-CNN P-ResNet P-ResNet
       with WP with UP with WP with UP
Simulated condition
REVERB SRMR 6.34 6.64 7.37 7.90 8.16 7.46 8.08 7.84
Eval LLR   0.57 0.49 0.47 0.79 0.53 0.48 0.49
  PESQ   2.27 2.57 2.68 2.16 2.43 2.73 2.57
Real condition
REVERB Eval SRMR 3.44 3.74 5.86 5.79 3.84 7.23 7.00 6.83
VoiceHome V0.2 SRMR 3.23 3.38 5.80 5.69 2.58 5.49 7.32 5.72
VoiceHome V1.0 SRMR 4.04 4.47 5.89 6.13 2.78 5.81 7.31 6.27
Average
AVG ±STD SRMR 4.26 ±1.24 4.56 ±1.26 6.23 ±0.66 6.38 ±0.89 4.34 ±2.26 6.50 ±0.86 7.43 ±0.40 6.66 ±0.78
  1. Bold results correspond with the best dataset value, and italic results show the second-best value