Can this be fine-tuned/trained on a new language entirely?

[Aryan185]

I want to use this model but for Indic languages. I did see the the training code mentioned in the other issue but that would fine-tune it to generate output in Indic format. Transcription input would still be an issue. Any help/suggestion would be appreciated.