Japanese

[4Sol]

I tried using WebnVTT adding Japanese, so I have both mixed English and JP subtitles, but it was working quite bad (JP) and it was not using kanji, but romaji instead. I don't see anywhere it could have used a model for JP at WebVTT fields, so.. what am I doing wrong?

Basically I want to speak in English and then in JP, so the subtitle catch both languages and put it in 1 sentence

[Tabby]

I don't think you're doing anything wrong here, it seems more likely that it just doesn't handle switching between languages very well. By default (IIRC) it attempts to auto-detect which language you're speaking in and doesn't ever change that once it's chosen one, and it sounds like it's picking English and then attempting to transcribe the Japanese as if it were English which would be why it wasn't working well and coming out as romaji. You can force it to be a particular language in the settings but then you'd be locking it into only one of those languages, so you'd have to switch the language in the settings when you want to change which is a rubbish user experience, even if you could bind hotkeys to do so

I'll need to do some research and figure out whether it's possible to tell Whisper to use multiple languages, or set things up in a way that makes it better at re-evaluating each chunk to guess at which language it is or something like that, but this might just be a use case that doesn't work terribly well