Speech Separation on VCTK 2 Speech
14.52SI-SDRTF-Locoformer
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| TF-LocoformerLearning Paradigm=Supervised2025.12 | 14.52 | 2.42 | — | |
| Conv-TasNetLearning Paradigm=Supervised2025.12 | 9.63 | 1.8 | — | |
| UnasDiffLearning Paradigm=Unsupervised, Strategy=Gradient+Initialization2025.12 | 6.11 | 1.62 | 29.8 | |
| DSG’s σ(t)-prop. γ(t)Learning Paradigm=Unsupervised, Strategy=Gradient+Initialization2025.12 | 5.68 | 1.57 | 31.3 | |
| Analytical SamplingLearning Paradigm=Unsupervised, Strategy=Analytical Sampling2025.12 | 2.17 | 1.28 | 42.3 | |
| DPS’s constant γ(t)Learning Paradigm=Unsupervised, Strategy=Gradient+Initialization2025.12 | 1.87 | 1.43 | 46.5 | |
| Unprocessed2025.12 | 0.01 | 1.29 | — |