Local AI Dictation

Local microphone dictation for Linux with:

• Whisper and Parakeet backends
• a localhost bridge
• a native Electrobun GUI
• Waybar / Hyprland integration

Default backend: Whisper.

Main commands

local-ai-dictation dictation
local-ai-dictation bridge
local-ai-dictation bridge-toggle
local-ai-dictation gui
local-ai-dictation backend get
local-ai-dictation backend set whisper
local-ai-dictation backend set parakeet
local-ai-dictation backend toggle --restart-bridge
local-ai-dictation devices --json
local-ai-dictation doctor --json
local-ai-dictation benchmark --fixture tests/fixtures/short_16k.wav --runs 2 --json

Backends

Whisper

• default
• lighter on VRAM
• current model: deepdml/faster-distil-whisper-large-v3.5

Parakeet

• heavier on VRAM
• current model: nvidia/parakeet-tdt-0.6b-v3

Install

uv python install 3.12
uv venv --python 3.12
source .venv/bin/activate

# GPU PyTorch
uv pip install torch torchaudio torchvision --index-url https://download.pytorch.org/whl/cu130

# install app
uv pip install -e .

System packages commonly needed on Linux:

• uv
• bun
• portaudio
• wl-clipboard or xclip
• desktop deps for Electrobun/WebKitGTK

Config

Config file:

~/.config/local-ai-dictation/config.toml

Example:

backend = "whisper"
cpu = false
input_device = ""
vad = false
max_silence_ms = 1200
min_speech_ms = 300
vad_mode = 2
format = "text"
output_file = ""
clipboard = true
debug = false

Supported env vars:

• LOCAL_AI_DICTATION_BACKEND
• LOCAL_AI_DICTATION_CPU
• LOCAL_AI_DICTATION_INPUT_DEVICE
• LOCAL_AI_DICTATION_VAD
• LOCAL_AI_DICTATION_MAX_SILENCE_MS
• LOCAL_AI_DICTATION_MIN_SPEECH_MS
• LOCAL_AI_DICTATION_VAD_MODE
• LOCAL_AI_DICTATION_FORMAT
• LOCAL_AI_DICTATION_OUTPUT_FILE
• LOCAL_AI_DICTATION_CLIPBOARD
• LOCAL_AI_DICTATION_DEBUG

GUI

Run:

local-ai-dictation gui

The GUI shows:

• current running model
• bridge command
• transcript history
• model switch button

Switching models from the GUI restarts the bridge live. The desktop GUI now subscribes to the bridge /events SSE stream for live state updates instead of polling bridge state on a timer.

Hyprland / Waybar

Current local integration uses:

• ~/.local/bin/local-ai-dictation-bridge
• ~/.config/hypr/scripts/local-ai-dictation-toggle.sh
• ~/.config/hypr/scripts/local-ai-dictation-switch-model.sh
• ~/.config/waybar/scripts/local-ai-dictation-status.py

Behavior:

• left click Waybar icon: start / stop recording
• right click Waybar icon: switch model and restart bridge
• Waybar shows: 󰍬 Whisper or 󰍬 Parakeet
• Waybar no longer polls bridge state on an interval. The bridge pushes RTMIN+8 to Waybar on state changes, and the module refreshes on startup and on that signal.
• The bridge still keeps /health for one-shot checks and recovery paths.

Hyprland startup currently launches the bridge with:

exec-once = /home/devkit/.local/bin/local-ai-dictation-bridge &

Notes

• On Linux Wayland, the desktop app still does not own a true global app hotkey yet. Ctrl+Alt+R works when the Local AI Dictation window is focused, but a compositor-level shortcut is still needed for truly global activation until GlobalShortcuts portal support is added.
• Switching backend clears the old model from VRAM by restarting the bridge process.
• Whisper is the safer backend on 6 GB cards.
• Parakeet currently has a larger load-time VRAM spike before settling.

Verification

python -m compileall src tests
.venv/bin/python -m pytest -q tests/test_bridge.py
cd desktop/electrobun && bun run check