Audio-text alignment correlation on AudioCaps (test)
0.457SRCCHuman-CLAP
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Human-CLAPLoss Function=wSCE + MAE2025.06 | 0.457 | 0.481 | 0.32 | 0.057 | |
| Human-CLAPLoss Function=MAE2025.06 | 0.453 | 0.472 | 0.317 | 0.051 | |
| Human-CLAPLoss Function=wSCE + MSE2025.06 | 0.447 | 0.471 | 0.313 | 0.059 | |
| Human-CLAPLoss Function=MSE2025.06 | 0.439 | 0.462 | 0.307 | 0.056 | |
| Human-CLAPLoss Function=wSCE2025.06 | 0.383 | 0.41 | 0.265 | 0.063 | |
| LAION CLAP2025.06 | 0.28 | 0.294 | 0.192 | 0.068 | |
| MS CLAP2025.06 | 0.278 | 0.296 | 0.192 | 0.078 |