Support for adding new languages

[assefvisic]

Hi Liquid team

Amazing work on Liquid Audio — I’ve been exploring it and really like the interleaved text/audio design.

What would be the recommended approach to add support for a new language, for example Croatian (hr)?

I’d like to understand:

• Which model components are language-dependent (text tokenizer, audio encoder, etc.)?
• Is fine-tuning LFM2-Audio on new audio↔text pairs sufficient, or would this require changes to the tokenizer or encoder?
• Any rough guidance on data requirements (hours or token count) for a usable new language model?

If you have any internal recipes, tips, or references for multilingual extensions, I’d really appreciate it.

Thanks a lot for your time and for releasing such an interesting system! 🙏

[auli-aziz]

Mark