Tripura ST Lab
Training code and dataset registry for Kokborok-first speech translation, then Bengali and Marathi on BhasaAnuvaad.
Landing page: https://huggingface.co/spaces/vivekharry/tripura-st-lab-site
Kokborok AUDIO (16 kHz mono)
→ ASR (MMS-1B + NE-ASR adapter, or Whisper fine-tune)
Kokborok transcript
→ MT (NLLB / sdmy/kokborok)
English
Clone this repo, then:
pip install -e .
python -m indic_st.train_asr --dataset kokborok_asr --init-from openai/whisper-small
python -m indic_st.train_mt --dataset kokborok_mt
See the full README in the files tab for dataset IDs, tonal-augmentation rules, and citations.
Author: Vivek Das (vivekharry).