Indic Speech-to-Text

Models

Download or remove the real transcription model weights. This only manages files on disk — to actually transcribe with them instead of the mock backend, set ASR_BACKEND=indicconformer, install the heavy Python dependencies from requirements.txt, and restart the backend.

IndicConformer is a gated model on Hugging Face — Install will fail until you log in, request/accept access on its model page, and set HF_TOKEN in backend/.env.