ElishaAz / Sayboard

An open-source on-device voice IME (keyboard) for Android using the Vosk library.
GNU General Public License v3.0
384 stars 23 forks source link

Punctuation models #14

Open sandoxy opened 1 year ago

sandoxy commented 1 year ago

There are punctuation models, which are quite big. Is there a possibility to use these models?

https://alphacephei.com/vosk/models

Screenshot_2022-12-22_10-07-27.png

ElishaAz commented 1 year ago

They seem to be targeting Desktop or Server. See https://github.com/benob/recasepunc/issues/14 - Most phones don't have that much RAM. Perhaps we can run it on a computer similar to https://github.com/ElishaAz/Sayboard/issues/10.

Anyways, it looks like they do the prediction off-line (i.e. after they have all the text) which is not practical with an IME.

nshmyrev commented 1 year ago

Its probably better to implement some kind of verbalized punctuation. i.e. when user says "comma" in the end it puts comma. In general we will try to implement lightweight mobile models, but it will take some time

sandoxy commented 1 year ago

when user says "comma" in the end it puts comma.

It would be great!

sandoxy commented 1 year ago

Most phones don't have that much RAM.

I thought, that this would be a problem.

savvyyyyy commented 1 year ago

Let me just say that you're onto something amazing. There's very few foss working options and even less that work by using just one app. Wondering if the punctuation buttons could be brought back like the ones in the initial release ? It'd help greatly. Or are they there and I'm just not working it out ? Thanks again