Self-supervised pretraining on Libri-Light 6k
23Pretraining Time (hr)SpidR
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SpidRGPUs=A100 ×16, Steps=400k, Batch size=63 min2025.12 | 23 | 369 | |
| data2vec 2.0GPUs=A100 ×16, Steps=50k, Batch size=17 min2025.12 | 43 | 688 | |
| HuBERTGPUs=A100 ×32, Steps=650k, Batch size=47 min2025.12 | 62 | 1,984 | |
| k2SSL ZipformerGPUs=V100 ×8, Steps=225k, Batch size=80 min2025.12 | 64 | 513 | |
| DinoSRGPUs=V100 ×16, Steps=400k, Batch size=63 min2025.12 | 180 | 2,880 | |
| Academic HuBERTGPUs=A100 ×8, Steps=1760k, Batch size=47 min2025.12 | 240 | 1,920 | |
| MelHuBERTGPUs=RTX 3090 ×1, Steps=630k, Batch size=8 min2025.12 | 300 | 300 |