Medium model produces nonsense

[tjongsma]

Hi there,

I've been using insanely-fast-whisper to try do some live streaming. I'm using this gist for reference: https://gist.github.com/Und3rf10w/a2561b569572677dd9c1f8db8e92fb25

Everything works fine for large-v3 (albeit somewhat slowly), but if I use the medium model the output becomes nonsense (funnily enough the output keeps saying 'I'm sorry I don't know what to do'.

Any idea on why this happens/why to fix it? Thank you!