Plume — offline voice dictation

Hold a shortcut, speak, release: Plume types what you said in the app you're using. Transcription runs locally, nothing is sent online.

Install

Download the latest version from Releases:

• macOS (Apple Silicon): .dmg
• Windows (x64): .exe installer
• Linux (x64, Ubuntu 24.04+, X11 only): .deb, install with sudo apt install ./Plume-*.deb

Builds aren't signed, so macOS or Windows may show a security warning.

Usage

On first launch, Plume asks you to pick a speech model and grant microphone access (plus Accessibility and Input Monitoring on macOS). It then lives in your menu bar or system tray.

Shortcut	Action
Ctrl + Space (hold)	Record while held, insert text on release
Ctrl + Shift + Space	Start / stop hands-free recording
Esc	Cancel

Shortcuts can be changed in Settings → Dictation. Recent transcriptions and the last 8 recordings are kept in history, so you can copy or retranscribe them.

Models

Model	Size	Languages
Whisper Base	142 MB	99
Whisper Small	466 MB	99
Parakeet TDT v3	639 MB	25 European
Nemotron 3.5 Compact	793 MB	35
Whisper Large-v3-Turbo	1.5 GB	99

Smaller models are faster, larger ones more accurate. The app currently supports French, English and automatic detection. Parakeet TDT v3 is NVIDIA's model (CC BY 4.0), via the ONNX conversion by Ilya Stupakov.

Privacy

Audio and history stay on your computer. Internet is only used to download models and check for updates.

Development

See the development guide. Bugs and ideas go in issues.