Automatic Speech Recognition on Competition Audio (test)
16.751WEROurs (tugstugi FT, beam=5)
Evaluation Results
| Method | Links | |
|---|---|---|
| Ours (tugstugi FT, beam=5)Params=769M, FT?=Yes, beam=52026.02 | 16.751 | |
| tugstugi + vocals + normalizationParams=769M, FT?=No, normalization=true, vocals=true2026.02 | 34.8 | |
| tugstugi + normalizationParams=769M, FT?=No, normalization=true2026.02 | 35.9 | |
| IndicWav2Vec BengaliParams=317M, FT?=No2026.02 | 40 | |
| BanglaASR small (5 ep.)Params=244M, FT?=Yes2026.02 | 50 | |
| Whisper large-v3Params=1,550M, FT?=No2026.02 | 75 |