Conditional Audio Generation on SC09 (test)
3.26MOS ScoreScore Subnet
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Score SubnetTotal Parameters (M)=12.3, Trainable Parameters (M)=4.4, GMACs=16.44, Guidance strength (gamma)=3.02026.06 | 3.26 | 1.02 | 0.03 | 8.65 | 287.19 | 15 | |
| U-Net + ClassifierTotal Parameters (M)=24.5, Trainable Parameters (M)=16.6, GMACs=22.74, Guidance strength (gamma)=3.02026.06 | 3.25 | 0.82 | 0.03 | 8.36 | 260.5 | 19 | |
| Score Subnet (no gradient taps)Total Parameters (M)=11.9, Trainable Parameters (M)=4.0, GMACs=11.50, Guidance strength (gamma)=3.02026.06 | 3.23 | 0.94 | 0.02 | 8.05 | 223.98 | 22 |