From: Shallow and deep feature fusion for digital audio tampering detection
Fusion method | Classical dataset | GAUDI-DI dataset | Dropout rate | ||
---|---|---|---|---|---|
Accuracy (%) | F1-score (%) | Accuracy (%) | F1-score (%) | ||
With attention | 95.05 | 95.41 | 87.92 | 87.71 | 0 |
96.04 | 96.12 | 87.78 | 87.46 | 0.1 | |
97.03 | 96.91 | 88.31 | 88.17 | 0.2 | |
96.04 | 96.12 | 87.65 | 87.32 | 0.3 | |
Without attention | 94.06 | 94.33 | 87.38 | 87.13 | 0 |
95.05 | 95.41 | 88.05 | 87.82 | 0.1 | |
95.05 | 95.41 | 87.92 | 87.75 | 0.2 | |
95.05 | 95.41 | 87.25 | 87.06 | 0.3 |