Feature: Character-level Syncing

[IlIllIl]

Hi,
Thank you for creating such an awesome tool.I've found that the tool cannot recognize languages without space separation as single words. I would be grateful if you could add character-level syncing for songs.

[pxeemo]

Thanks for the kind words! Just to clarify—are you referring to CJK (Chinese, Japanese, Korean) languages? If possible, could you share an example or sample of a line in LRC format? That would help a lot!

[pxeemo]

I looked into this a bit. It seems better to use proper tokenization to separate words instead of treating every character individually. However, that can be tricky—it might require running NLP tools on a server or using cloud-based APIs like Google or Azure which have their limits.

In the meantime, you can ask an LLM to insert <> between words manually as a workaround.

[IlIllIl]

Thank you for your reply. I use regular expressions to filter single characters of CJK and manually add <> now. It's better if the tool can split CJK characters without extra steps. Technically, external NLP tools is needed for filtering out CJK phrases and idioms, but I think single characters is enough.