Does this model support Thai and English code-switching?

#1
by FILM6912 - opened

Hi! Thanks for releasing this model.

I have a few questions about multilingual support.

  1. Does this model officially support both Thai and English?

  2. Can it recognize code-switching (Thai and English mixed in the same sentence)? For example:

    วันนี้เราจะ review the project before deployment.

  3. If mixed-language transcription is supported, is there any recommended configuration (e.g., language setting, tokenizer, decoder prompt, or other inference parameters) to achieve the best results?

  4. Or should the language be set to auto instead of forcing th or en?

Thanks in advance!

Thanks for your interest!
This model officially supports Thai only. It can handle code-switching at the word level (individual English words mixed into Thai speech), but it was fine-tuned exclusively on a Thai dataset, so the English capability inherited from the original Nemotron model is expected to be degraded.

Sign up or log in to comment