The offline AI that I tried a few months ago probably needed training on the noise environment to get decent results. I forget which ones I tried but likely the ones with extensions already in Oobabooga Textgen. I was messing with text to speech mostly, but some of the ones that do TTS also have packages and examples for STT. Nothing I tried for offline generation was good enough to speak as an AI prompt without manual corrections.
Check out Talon Voice. It’s a bit of getting your head around, but I use it for not just transcribing, but also coding, and general computer navigation.
But basically look for something that utilizes the Whisper AI model. This list has a few options, but I am not sure if there is newer stuff: github.com/sindresorhus/awesome-whisper
I can’t believe how good this is! I’m on a really busy street (in the middle of Sydney, George Street actually), and it’s put in commas in, it handles question marks and all my texts. I can’t believe how good this system is and to think that it is local and open source is so impressive.
I know, right? It’s a really amazing piece of work when it comes right down to it. They have a donation thing, but since I don’t have Google Play Services, I can’t use it, and they don’t have the crypto donation set up yet.
Thanks for letting me know about the donation thing. I suffer from RSI so I’m always happy to donate to things that help. Much appreciated. And as before, this is written with the keyboard voice to text. It’s now very, very noisy in a bar.
Add comment