buzz?
github.com/chidiwilliams/buzz/

@cachondo looks like it'll work at least, yeah. thx

I use it for generating subtitles for the internal videos I produce. I have to edit them for general hallucinatory content as with all these tools, but it's still quicker than typing it out by hand.

@cachondo If You use the the larger model, the accuracy issue is almost resolved. You also have to use NVDA's object nav to use some of the interface.