Text-to-Audio on AudioCaps 2019 (test)
1.558FADUNISON (D24, 16kHz)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| UNISON (D24, 16kHz)Architecture=MM-DiT, Params=732M, Variant=D24, Sampling Rate=16kHz2026.05 | 1.558 | 16.28 | 1.459 | 10.9 | 50.3 | |
| GenAU-LArchitecture=DiT, Params=1.25B, substantially larger training data or preference optimization=true2026.05 | 1.591 | 18.41 | 1.29 | 11.94 | 56.1 | |
| UNISON (D20, 44kHz)Architecture=MM-DiT, Params=621M, Variant=D20, Sampling Rate=44kHz2026.05 | 1.756 | 15.82 | 1.455 | 12.04 | 46.7 | |
| TangoArchitecture=UNet, Params=866M2026.05 | 1.846 | 24.52 | 1.305 | 7.45 | 49.8 | |
| Make-An-Audio 2Architecture=DiT, Params=937M2026.05 | 2.142 | 20.14 | 1.597 | 10.02 | 44.1 | |
| Audio-OmniArchitecture=DiT, Params=3.05B2026.05 | 2.535 | 31.42 | 1.337 | 9.55 | 48.6 | |
| AudioLDM 2-LargeArchitecture=UNet, Params=712M2026.05 | 3.097 | 29.68 | 1.49 | 7.98 | 45.2 | |
| UniSonateArchitecture=MM-DiT, Params=1.34B2026.05 | 4.21 | 30.21 | 2.44 | 8.22 | — | |
| MMAudio-LArchitecture=MM-DiT, Params=1.03B2026.05 | 5.893 | 16.53 | 1.421 | 11.98 | 44.1 | |
| Stable Audio OpenArchitecture=DiT, Params=1.06B2026.05 | 10.83 | 52.03 | 3.049 | 6.13 | 20.3 |