Video-to-Audio Generation on AudioCanvas (out-of-domain)
52CLAPPrismAudio
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| PrismAudiotype=Ours2025.11 | 52 | 36 | 6.68 | 2.82 | 4.26 | 6.15 | 3.5 | 12.87 | 1.92 | 1.53 | 4.12 | 4.01 | |
| GTtype=Ground Truth2025.11 | 48 | 40 | 6.47 | 3.16 | 4.02 | 5.99 | — | — | — | — | 4.65 | 4.72 | |
| ThinkSoundtype=Baseline2025.11 | 48 | 80 | 6.48 | 3.5 | 4.1 | 5.94 | 4.43 | 22.82 | 1.95 | 2.54 | 3.79 | 3.8 | |
| PrismAudiovariant=Silent Video CoT2025.11 | 47 | 42 | 6.55 | 3.04 | 4.09 | 5.98 | 3.63 | 14.26 | 2.01 | 1.79 | — | — | |
| MMAudiotype=Baseline2025.11 | 46 | 43 | 6.3 | 3.23 | 3.97 | 5.77 | — | — | 3.59 | 1.87 | 3.88 | 3.87 | |
| HunyuanVideo-Foleytype=Baseline2025.11 | 44 | 47 | 6.43 | 3.25 | 4.04 | 5.88 | — | — | 2.04 | 2.07 | 3.75 | 3.71 | |
| PrismAudiovariant=w/o CoT-RL2025.11 | 42 | 44 | 6.45 | 3.22 | 3.81 | 5.87 | 4.11 | 15.3 | 2.1 | 2.17 | 3.91 | 3.85 |