Voice Dictation (Fedora/Wayland)

Continuous voice dictation using Silero VAD + Whisper + ydotool.

How it works

1. Press Super+D → Start continuous listening
2. Speak your text
3. Pause ~1.8 seconds → Automatic transcription
4. Text appears at cursor position
5. Keep speaking (repeats automatically)
6. Say "enter" at end of a sentence → sends Enter key
7. Press Super+D → Stop

Requirements

System packages

sudo dnf install alsa-utils ydotool libnotify

User must be in the input group for ydotool:

sudo usermod -aG input $USER
# Log out and back in

Python packages

cd ~/Projects/Personal/dictation
uv sync

This creates a .venv/ with all dependencies. After a Python version upgrade (e.g. Fedora upgrade), rebuild with:

bin/rebuild-venv.sh

Installation

The scripts live in bin/ and are symlinked to ~/.local/bin/:

ln -sf ~/Projects/Personal/dictation/bin/dictation-continuous ~/.local/bin/dictation-continuous
ln -sf ~/Projects/Personal/dictation/bin/dictation-toggle ~/.local/bin/dictation-toggle

GNOME keybinding

Set up via GNOME Settings → Keyboard → Custom Shortcuts:

• Name: Toggle Dictation
• Command: ~/.local/bin/dictation-toggle nl
• Shortcut: Super+D

Configuration

Edit bin/dictation-continuous to adjust:

Parameter	Default	Description
SILENCE_THRESHOLD	1.8	Seconds of silence before processing
VAD_THRESHOLD	0.45	VAD sensitivity 0–1 (higher = less sensitive)
MIN_SPEECH_DURATION	0.5	Minimum speech duration in seconds

Debugging

Logs are written to /tmp/dictation-debug.log:

tail -f /tmp/dictation-debug.log

Known issues

1. PipeWire/PulseAudio piping doesn't work — uses arecord (direct ALSA) instead
2. wtype doesn't work on GNOME Wayland — uses ydotool instead
3. First start is slow — VAD and Whisper models need to load