Table 5 Experimental results of combining different modules with MSTCM
From: Supervised Attention Multi-Scale Temporal Convolutional Network for monaural speech enhancement
| Â | #Param. (M) | MACs (G/s) | PESQ | STOI (%) | SDR |
|---|---|---|---|---|---|
Unprocessed | - | - | 2.24 | 88.5 | 14.81 |
MSTCMs | 10.13 | 1.02 | 2.99 | 92.3 | 19.08 |
 +U\(^2\)-LSTM | 23.54 | 8.26 | 3.32 | 93.9 | 20.16 |
  +Supervised Attention | 23.55 | 8.34 | 3.39 | 94.1 | 20.45 |
   +GTCMs | 24.66 | 8.45 | 3.44 | 94.3 | 20.61 |
    +Compensation | 28.12 | 12.07 | 3.47 | 94.3 | 20.83 |
MSTCMs | 10.13 | 1.02 | 2.99 | 92.3 | 19.08 |
 +GTCMs | 11.24 | 1.13 | 3.16 | 92.9 | 19.75 |
  +U\(^2\)-LSTM | 24.65 | 8.37 | 3.37 | 94.1 | 20.43 |
   +Supervised Attention | 24.66 | 8.45 | 3.44 | 94.3 | 20.61 |
    +Compensation | 28.12 | 12.07 | 3.47 | 94.3 | 20.83 |