Audio-to-Text Retrieval on MusicBench (test)
39.09Recall@1FIGMA
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| FIGMA2026.06 | 39.09 | 68.02 | 80.71 | 88.83 | |
| LAION-CLAPTraining Data=General Audio, Evaluation Protocol=Zero-shot2026.06 | 27.92 | 53.81 | 76.65 | 86.8 | |
| LAION-CLAPTraining Data=Music, Speech and General Audio, Evaluation Protocol=Zero-shot2026.06 | 26.9 | 59.39 | 71.57 | 83.76 | |
| LAION-CLAPTraining Data=Music, Evaluation Protocol=Zero-shot2026.06 | 25.38 | 61.93 | 76.14 | 89.34 | |
| LAION-CLAPTraining Data=General audio with variable-length, Evaluation Protocol=Zero-shot2026.06 | 23.35 | 52.79 | 62.94 | 78.17 | |
| LAION-CLAPTraining Data=Music and Speech, Evaluation Protocol=Zero-shot2026.06 | 19.8 | 48.22 | 65.48 | 80.2 | |
| LAION-CLAPNote=Continued Training on FGMCaps2026.06 | 13.71 | 36.55 | 52.79 | 70.05 |