TinyWhisper

Note: This project is no longer actively maintained.

Local voice-to-text for macOS Apple Silicon. Press a hotkey, speak, and the transcription is pasted at your cursor.

Runs entirely on-device using MLX — no cloud APIs, no subscriptions.

Screenshots

Welcome & Setup	Advanced Settings	Waveform Overlay

Features

• Fast local transcription — Parakeet TDT 0.6b via MLX (~70x realtime)
• Global hotkey — Option+Space to toggle recording (rebindable)
• Auto-paste — transcribed text goes straight to your cursor
• Waveform overlay — shows recording status without stealing focus
• Configurable — colors, opacity, hotkey, overlay size via settings UI or YAML
• System tray — lives in your menu bar with live status, memory usage, and quick access to settings

Requirements

• macOS on Apple Silicon (M1/M2/M3/M4)
• Python 3.10+
• uv (recommended — manages Python and dependencies)

Install

uv sync

To also install Whisper support:

uv sync --extra whisper

macOS .app bundle

Build a native menu-bar app (recommended — gives TinyWhisper its own identity for macOS permissions):

uv sync --no-editable   # required — editable installs trigger a Documents TCC prompt
uv run build_app.py
cp -r TinyWhisper.app /Applications/   # optional
open TinyWhisper.app

Or use the all-in-one setup script to install, build, and launch with animated progress:

./setup.sh

Usage

uv run tinywhisper     # run from terminal
# or
open TinyWhisper.app   # run as native app

1. Grant Input Monitoring, Accessibility, and Microphone permissions when prompted
2. Press Option+Space to start recording
3. Speak, then press Option+Space again
4. Transcription is pasted at your cursor

Right-click the menu bar icon for settings, audio device selection, config file, and status info.

Configuration

Settings are stored at ~/.config/tinywhisper/config.yaml. You can edit via the tray menu (Settings or Open Config File) or manually:

hotkey:
  modifier: "option"       # option, ctrl, cmd, shift
  key: "space"             # space, tab, f1-f12, etc.

transcription:
  engine: "parakeet"       # or "whisper"
  parakeet:
    model: "mlx-community/parakeet-tdt-0.6b-v3"
  whisper:
    model: "mlx-community/whisper-large-v3-turbo"

recording:
  sample_rate: 16000
  channels: 1
  device: null             # audio input device name, or null for system default

overlay:
  enabled: true
  width: 300
  height: 60
  opacity: 0.85
  color: "#FF6B6B"
  bg_color: "#1E1E1E"

Transcription Engines

Engine	Model	Speed	Memory
parakeet (default)	Parakeet TDT 0.6b v3	~70x realtime	~600 MB
whisper	Whisper Large V3 Turbo	~15-20x realtime	~1.5 GB

Models are downloaded from HuggingFace on first run and cached locally.

License

MIT