Arch Linux

5hridhyan

Member

Registered: 2025-12-25

Posts: 1,098

Always-on voice trigger for executing system command(hyprland)

I want to build a minimal voice-triggered command system for my Arch setup (Hyprland)...

my current goal:
- similar to Bixby/voice assistants, but without any responses/talkback, just does what I/user says/said
- just detect a trigger phrase (smtg like “hey meow”) and execute simple commands like:
  - open browser (if possible, search " tom and jerry on Youtube" etc )
  - suspend / poweroff
  - launch or close applications
  -etc

So effectively:
continuous listening => detect trigger => parse command => execute shell command

I’m kinda unsure/konfused about:
- is continuous speech-to-text + trigger phrase detection practical on a lightweight system, or too resource-heavy? if heavy, implement wake word and then load STT?
- how are always-on audio listeners typically implemented on Linux (daemon, service, etc.)?
- for Wayland/Hyprland, is it better to rely purely on shell commands, or use compositor IPC for actions?

I have no plans on/to using heavy assistant frameworks, KISS (keeping it simple and smart)....

Last edited by 5hridhyan (2026-03-27 16:30:23)

Ruhanilinks

Member

Registered: 2026-09-11

Posts: 22

Re: Always-on voice trigger for executing system command(hyprland)

I think your KISS approach makes sense for this. You probably don't need continuous full speech-to-text running all the time. A lightweight wake-word detector could listen continuously, then start the STT process only after the trigger phrase is detected.

For the actual actions, I'd keep the command layer simple and use normal shell commands where possible. Hyprland's IPC is useful for things that are specifically compositor-related, such as switching workspaces, moving windows, or managing clients.

Running the listener as a small user systemd service also seems like a clean fit. That would let it start with your session and keep the voice component separate from the command execution. The main thing I'd be careful about is restricting which commands the speech parser is allowed to execute rather than passing arbitrary recognized text directly to a shell.