Bengali speech-to-text

Two 300M CTC models, one API. Use the API link at the bottom of this page to call it from your project.

device cuda · accelerator zero · input cap 300s · 5-gram LM off

  • XLS-R 300M (arijitx) — ready · arijitx/wav2vec2-xls-r-300m-bengali
  • IndicWav2Vec v1 (ai4bharat) — ready · ai4bharat/indicwav2vec_v1_bengali

5-gram KenLM decoding is off (optional dependency missing: pyctcdecode).
Acoustic model