Speech Model updates for VOSK or Whisper

[loninappleton]

Whether it's VOSK or Whisper, how is anyone to know when the model is updated and improved for accuracy?

[Cornucopia]

Download the database then, and download it now (same level/version) and then do a bitwise compare. Or file timestamp compares.
Or read any accompanying changelogs.

That's about it.

Whisper sounds VERY interesting...if you have lots of time to spare/wait, and you are using English language. Otherwise, still not ready for prime time (too high error rates, holes in model/stats logic)

Scott

[loninappleton]

The models are often held though (at least with Subtitle Edit) in uncommon file locations. At least so-called roaming locations are obscure to me.
Where I personally alter subtitles made with the AI tools, it's still by hand. I have only done a couple of plays from start to finish as a hobby for some favorites.

Do the models have a specific file extension? Using SE makes any of this obscure since loading the models is inside SE. It's a great convenience but
doing these sorts of things discussed requires more detail. I use the Everything file search tool. That works well for locating things.

Ok, I did a practice looking up VOSK and Everything found it. VOSK does have a version number .22 on it's zip file. Whisper seems to go by date stamp--
last one 22 11 16.