Degraded Frame Detection on LibriAugmented clean-partial-mixup (test)
92AUC (I0.7,0.7)CON1
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CON1Detection type=Embedding-based, Input representation=zscl, Training loss=with Lscl2026.05 | 92 | 4.88 | 0.62 | |
| CON1Detection type=Embedding-based, Input representation=z-tilde, Training loss=with Lscl2026.05 | 92 | 4.64 | 0.61 | |
| CON1Detection type=Embedding-based, Input representation=x, Training loss=with Lscl2026.05 | 86 | 7.32 | 1 | |
| CON2Detection type=Embedding-based, Input representation=zscl, Training loss=with Lscl2026.05 | 80 | 7.6 | 0.95 | |
| SUP2Detection type=Embedding-based, Input representation=x, Training loss=without Lscl2026.05 | 79 | 11.07 | 0.95 | |
| CON2Detection type=Embedding-based, Input representation=x, Training loss=with Lscl2026.05 | 78 | 9.51 | 1 | |
| CON2Detection type=MOS-based, Input representation=z-tilde, Training loss=with Lscl2026.05 | 78 | — | — | |
| CON1Detection type=MOS-based, Input representation=z-tilde, Training loss=with Lscl2026.05 | 77 | — | — | |
| SUP2Detection type=MOS-based, Input representation=zMOS, Training loss=without Lscl2026.05 | 76 | — | — | |
| SUP2Detection type=Embedding-based, Input representation=zMOS, Training loss=without Lscl2026.05 | 72 | 10.79 | 1 | |
| SUP1Detection type=Embedding-based, Input representation=x, Training loss=without Lscl2026.05 | 63 | 13.34 | 0.93 | |
| SUP1Detection type=MOS-based, Input representation=zMOS, Training loss=without Lscl2026.05 | 59 | — | — | |
| SUP1Detection type=Embedding-based, Input representation=zMOS, Training loss=without Lscl2026.05 | 43 | 12.16 | 0.93 | |
| BaselineDetection type=Embedding-based, Input representation=x, Training loss=without Lscl2026.05 | 30 | 15.29 | 1 | |
| BaselineDetection type=MOS-based, Input representation=zMOS, Training loss=without Lscl2026.05 | 20 | — | — | |
| BaselineDetection type=Embedding-based, Input representation=zMOS, Training loss=without Lscl2026.05 | 7 | 16.61 | 1 | |
| CON2Detection type=Embedding-based, Input representation=z-tilde, Training loss=with Lscl2026.05 | 5 | 26.7 | 1 |