Ambient Sound Generation on AVSync15 (test)
0.56DeSyncSyncDPO
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| SyncDPOTraining Data=ID2026.05 | 0.56 | 0.33 | 0.5 | 0.25 | 41.68 | 423.8 | 388.17 | 5.78 | |
| SyncDPOTraining Data=OOD2026.05 | 0.67 | 0.34 | 0.49 | 0.25 | 40.49 | 427.92 | 382.99 | 5.76 | |
| SFTTraining Data=ID2026.05 | 0.71 | 0.33 | 0.51 | 0.26 | 41.91 | 439.3 | 407.79 | 5.81 | |
| DPOTraining Data=ID2026.05 | 0.72 | 0.33 | 0.49 | 0.26 | 43.38 | 421.53 | 412.12 | 6.64 | |
| DPOTraining Data=OOD2026.05 | 0.75 | 0.33 | 0.49 | 0.26 | 43.44 | 419.01 | 397.81 | 6.24 | |
| No TuneTraining Data=None2026.05 | 0.81 | 0.33 | 0.48 | 0.26 | 43.31 | 422.88 | 388.31 | 6.03 | |
| SFTTraining Data=OOD2026.05 | 0.92 | 0.34 | 0.49 | 0.25 | 39.19 | 421.54 | 394.6 | 5.93 |