Text-to-Music Generation on 100 Official Final Prompts (test)
0.531ms-CLAP ScoreMusicGen-small
Evaluation Results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MusicGen-smallParams=300M, Data=∼20kh2026.05 | 0.531 | 6.86 | 7.47 | 5.22 | 7.4 | 7.4 | 8.36 | 8.93 | 5.58 | 5.44 | 7.05 | 3.6 | 3.55 | 3.55 | |
| MusicGen-largeParams=3.3B, Data=∼20kh2026.05 | 0.531 | 7.12 | 7.69 | 5.42 | 7.58 | 7.38 | 8.4 | 8.95 | 5.95 | 5.72 | 7.43 | 3.85 | 3.75 | 3.55 | |
| MusicGen-mediumParams=1.5B, Data=∼20kh2026.05 | 0.528 | 6.85 | 7.53 | 5.25 | 7.44 | 7.42 | 8.42 | 8.92 | 5.85 | 5.72 | 7.2 | 3.85 | 3.75 | 3.8 | |
| Stable Audio OpenParams=1.2B, Data=∼7.3kh2026.05 | 0.507 | 6.45 | 7.54 | 4.33 | 7.63 | 7.34 | 8.31 | 8.89 | 4.24 | 4.96 | 4.88 | 3.6 | 3.85 | 3.35 | |
| Efficiency: without lyric encoder (inference)Params=499M, Data=457h2026.05 | 0.5005 | 6.66 | 7.39 | 4.81 | 7.26 | 7.35 | 8.38 | 8.88 | 3.07 | 3.12 | 3.8 | 3.15 | 3.3 | 2.9 | |
| Efficiency: without timbre encoder (inference)Params=499M, Data=457h2026.05 | 0.4996 | 6.81 | 7.42 | 4.96 | 7.33 | 7.39 | 8.37 | 8.95 | 3.07 | 3.01 | 3.86 | 2.95 | 2.9 | 2.75 | |
| Efficiency Original, 16-layer DiTParams=499M, Data=457h2026.05 | 0.4993 | 6.92 | 7.4 | 5.17 | 7.33 | 7.41 | 8.4 | 8.89 | 3.08 | 2.9 | 4 | 2.95 | 3.1 | 2.75 | |
| retrained without lyric & timbre encoders, 23-layer DiTParams=489M, Data=457h2026.05 | 0.4844 | 6.5 | 7.11 | 5.02 | 7.05 | 7.35 | 8.3 | 8.92 | 2.6 | 2.6 | 3.28 | 2.75 | 2.9 | 2.65 | |
| Performance: without timbre encoder (inference)Params=2.4B, Data=457h2026.05 | 0.4826 | 6.53 | 7.31 | 4.7 | 7.25 | 7.37 | 8.37 | 8.9 | 3.55 | 3.72 | 4.32 | 3.15 | 3.25 | 3.25 | |
| retrained without lyric & timbre encoders, 24-layer DiTParams=1.8B, Data=457h2026.05 | 0.4776 | 6.47 | 7.06 | 5.2 | 7.01 | 7.4 | 8.39 | 8.95 | 2.03 | 2 | 2.81 | 2.25 | 2.55 | 2.25 | |
| Performance: without lyric encoder (inference)Params=2.4B, Data=457h2026.05 | 0.4739 | 6.38 | 7.24 | 4.6 | 7.15 | 7.35 | 8.35 | 8.87 | 3.44 | 3.63 | 4.36 | 3.1 | 3.4 | 3.1 | |
| Performance Original, 24-layer DiTParams=2.4B, Data=457h2026.05 | 0.4737 | 6.65 | 7.37 | 4.89 | 7.29 | 7.37 | 8.35 | 8.87 | 3.34 | 3.43 | 4.25 | 2.95 | 3.2 | 2.85 | |
| retrained without lyric & timbre encoders, 16-layer DiTParams=371M, Data=457h2026.05 | 0.4718 | 6.45 | 7.03 | 5.15 | 7.02 | 7.38 | 8.33 | 8.91 | 1.94 | 1.97 | 2.51 | 2.45 | 2.5 | 2.15 |