Automatic Speech Recognition on Bengali YouTube Competition (public)
16.751WERFull FT, A100, 12 epochs
Evaluation results
| Method | Links | |
|---|---|---|
| Full FT, A100, 12 epochsFine-tuning=Full FT, Hardware=A100, Training Epochs=12 epochs2026.02 | 16.751 | |
| + vocal extraction (HTDemucs)Vocal Extraction=HTDemucs2026.02 | 21 | |
| 50% data, H200, 5 epochsTraining Data=50%, Hardware=H200, Training Epochs=5 epochs2026.02 | 21.893 | |
| tugstugi base (no FT)Fine-tuning=no FT2026.02 | 34.8 |