Video-to-spatial audio generation on Hybrid (test)
4.6MOS (Subjective Quality)Ground Truth
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Ground Truth2026.05 | 4.6 | — | — | — | — | — | 4.58 | |
| SwanSphereParams=1.09B, Inf. Time=0.21s/9.13s2026.05 | 4.32 | 120.28 | 1.36 | 1.14 | 0.4 | 1.03 | 4.44 | |
| OmniAudioParams=1.22B, Inf. Time=0.85s2026.05 | 4.12 | 157.67 | 1.93 | 1.25 | 0.47 | 1.27 | 4.27 | |
| MMAudio+ASParams=1.03B, Inf. Time=2.76s2026.05 | 3.91 | 261.65 | 2.43 | — | — | — | 3.6 | |
| ViSAGeParams=0.36B, Inf. Time=20.19s2026.05 | 3.82 | 232.17 | 2.67 | 1.57 | 0.63 | 1.59 | 3.78 | |
| Diff-Foley+ASParams=0.94B, Inf. Time=2.03s2026.05 | 3.68 | 304.03 | 3.12 | — | — | — | 3.26 |