Numen is voice control for handsfree computing, letting you type efficiently by saying syllables and literal words. It works system-wide on Linux and the speech recognition runs locally.
There's a short demonstration on: numenvoice.org
Packages of Numen are available on:
and potentially other platforms.
scdoc are required.
A binary of the speech recognition library and an English model (about 40MB) can be installed with:
./get-vosk.sh && sudo ./get-vosk.sh install
./get-model.sh && sudo ./get-model.sh install
The dotool command which simulates the input, can be installed with:
./get-dotool.sh && sudo ./get-dotool.sh install
Finally, Numen itself can be installed with:
./build.sh && sudo ./build.sh install
dotool requires permission to
/dev/uinput to create the virtual input
devices, and a udev rule grants this to users in group input.
You can try:
echo type hello | dotool
and if need be, you could add your user to group input with:
sudo groupadd -f input
sudo usermod -a -G input $USER
and re-login and trigger the udev rule or just reboot.
If it types something other than hello, see about keyboard layouts in the manpage.
Once you've got a microphone, you can run it with:
There shouldn't be any output, but you can try typing hey by saying "hoof each yank", and try transcribing a sentence after saying "scribe". Terminate it by pressing Ctrl+c (aka "troy cap").
If nothing happened, check it's using the right audio device with:
timeout 5 numen --verbose --audiolog=me.wav
and specify a
--list-mics if not.
Now you're ready to have a go in your text editor! The default phrases are
I use Numen and the default phrases for all my computing, with keyboard-based programs like Neovim and qutebrowser. I also use a minimal desktop environment I made, called Tiles, that doesn't require a pointer device for window management, file picking, etc.
If you'd like to tweak the phrases, copy the default phrases to
~/.config/numen/phrases and edit them there. The manpage
You can send questions or patches by composing an email to ~firstname.lastname@example.org.
You're also welcome to join the Matrix chat at #numen:matrix.org.
AGPLv3 only, see LICENSE.
Copyright (c) 2022-2024 John Gebbie
You can append these lines to your model's conf/model.conf to make things extra responsive:
The default model is /usr/share/vosk-models/small-en-us, but you can edit a copy instead and specify it with the NUMEN_MODEL environment variable. We should be able to implement better with the next Vosk release.