Follow

github.com/ideasman42/nerd-dic # nerd dictation is a great off line speech to text program that works pretty well under linux and doesn't send your information out to the cloud. i even used it to type this very description accurately.

@climagic seems better than julius with the 40MB model already...

Though i bubblewrapped it and made xdotool my own script so i could see what it's doing.

Note that you can see the options using `nerd-dictation begin --help`, which wasnt immediately obvious to me..

@climagic looks like you can also get github.com/alphacep/vosk-api/p multiple results with confidences. Though you don't get those on the partials results.

Sometimes the context might be combine-able to get the right word.. Especially when you're looking for a command. Though probably i should grab a better model before trying that..

*note: you have to edit the source code to do that of course! 

Basically below `rec = vosk.KaldiRecognizer` set it to povide words and alternatives.

rec.SetMaxAlternatives(10)
rec.SetWords(True)

And then use `json_data` below `rec.AcceptWaveform` differently. (anything regarding `json_data["text"]` breaks, since the field is no longer in there...)

Sign in to participate in the conversation
Mastodon

Server run by the main developers of the project 🐘 It is not focused on any particular niche interest - everyone is welcome as long as you follow our code of conduct!