voxd 0.1.0

Voice dictation daemon for Linux with Sarvam AI STT

voxd

Voice dictation daemon for Linux with modular architecture:

• core: recording, Sarvam AI transcription, and clipboard integration
• wrapper.dwm: X11-focused wrapper for dwm
• wrapper.hyprland: Wayland-focused wrapper for hyprland

Press keybind once to start recording, press again to stop. Transcribed text is copied to clipboard.

Install

pip install voxd

System dependencies:

• ffmpeg (audio capture)
• xclip or xsel (clipboard on X11)
• wl-clipboard (clipboard on Wayland)

Setup

1. Configure API key

voxd config set api_key YOUR_SARVAM_API_KEY

2. (Optional) Set language

voxd config set language hi-IN  # Hindi
voxd config set language en-IN  # English (default)

3. Start daemon at boot

dwm / X11 - Add to ~/.xinitrc (before exec dwm):

voxd-daemon &

hyprland / Wayland - Add to ~/.config/hypr/hyprland.conf:

exec-once = voxd-daemon

4. Setup keybind

dwm - Edit your config.h:

{ MODKEY, XK_semicolon, spawn, SHCMD("voxd-dwm toggle") },

Then recompile: sudo make clean install

hyprland - Add to ~/.config/hypr/hyprland.conf:

bind = SUPER, semicolon, exec, voxd-hypr toggle

Reload: hyprctl reload

Usage

Press your keybind once to start recording, press again to stop. The transcribed text is copied to clipboard - paste with Ctrl+V!

Terminal commands:

voxd toggle              # Toggle recording
voxd status              # Check status
voxd quit                # Stop daemon
voxd config list         # Show config
voxd config set key val  # Set config value

Configuration

Config is stored in ~/.config/voxd/config.json

Available settings:

• api_key - Sarvam AI API key
• model - STT model (default: saaras:v3)
• language - Language code (default: en-IN)
• output_mode - Output mode (auto/x11/wayland)

Architecture

src/voxd/
  core/
    config.py          # env + runtime paths
    config_manager.py  # persistent config
    recorder.py        # ffmpeg process lifecycle
    sarvam_client.py   # Sarvam SDK integration
    injector.py        # clipboard integration
    service.py         # unix socket daemon + toggle logic
  wrappers/
    dwm.py             # X11 wrapper (output mode x11)
    hyprland.py        # Wayland wrapper (output mode wayland)
  cli.py               # daemon server + client commands

Notes

• Socket path: ${XDG_RUNTIME_DIR}/voxd/control.sock
• Captured .wav files: ${XDG_RUNTIME_DIR}/voxd/
• Status file: ${XDG_RUNTIME_DIR}/voxd/status
• Notifications show recording state via notify-send