Universal Sound Separation on dnr v2 (test)
6.8Music SI-SDRBiModalSS
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| BiModalSSEvaluation Protocol=zero-shot2025.09 | 6.8 | 1.8 | 6.36 | 2.5 | 2.6 | 2.3 | |
| Sudo rm-rf + FiLMData Training=dnr-v2, Evaluation Protocol=retrained2025.09 | 6.7 | 2 | 6.6 | 2.7 | 2.9 | 2.3 | |
| AudioSepData Training=dnr-v2, Evaluation Protocol=retrained2025.09 | 5.6 | 7.7 | 4.7 | 2.6 | 2.5 | 2.3 | |
| AudioSepEvaluation Protocol=zero-shot2025.09 | 2.5 | 4.9 | 0.3 | 2.9 | 3.1 | 2.6 | |
| CodecSepData Training=dnr-v2, Evaluation Protocol=retrained2025.09 | 1.2 | 10 | 0.9 | 2.9 | 3.2 | 2.3 | |
| CodecSepData Training=dnr-v2, Bitstream Configuration=codes in : codes out, Evaluation Protocol=zero-shot2025.09 | 0.2 | 8.3 | 1 | 2.5 | 3 | 2.3 | |
| CodecSepTraining Dataset=AudioCaps, Evaluation Protocol=zero-shot2025.09 | -8.5 | 2.5 | -5.9 | 2.3 | 2.6 | 2.1 | |
| AudioSepTraining Dataset=AudioCaps, Evaluation Protocol=zero-shot2025.09 | -14.9 | -7.1 | -14.6 | 2.4 | 2.4 | 2.2 |