Dictation (Speech-to-Text)
Dictate instead of type. Scriptorium’s dictation runs entirely on your device — your voice is never sent to any server.
Choosing an engine
Section titled “Choosing an engine”In Settings → AI Plugins, pick a dictation engine:
| Engine | Languages | Download |
|---|---|---|
| Disabled | — | none |
| Parakeet v2 | English | ~630 MB |
| Parakeet v3 | Multilingual | ~640 MB |
The Parakeet models are one-time downloads managed right in settings — install, and uninstall when you no longer need them.
Disabled is the default: no model, and the microphone button is off. Your operating system’s own dictation still works on top of Scriptorium, the way it does in any app — double-tap Fn on macOS — but that is the OS talking, not a Scriptorium engine, and none of the smart transcription below applies to it.
Dictating
Section titled “Dictating”- Place your cursor in a manuscript or note.
- Hold the push-to-talk shortcut — Ctrl+Space by default (configurable in settings).
- Speak. A floating waveform shows that you’re being heard.
- Release the shortcut — your words appear at the cursor.
Smart transcription
Section titled “Smart transcription”Dictation doesn’t just transcribe — it cleans up as it goes:
- Punctuation added automatically
- Filler words (“um”, “uh”) removed
- Pauses become paragraph breaks
- Your vocabulary — add proper nouns per project (character names, invented places) so they’re recognized correctly
- Custom replacements — define your own spoken phrases, e.g. saying “symbol open quote” inserts
"
Privacy
Section titled “Privacy”Audio is processed locally by the model on your machine. Nothing is uploaded, and dictation works fully offline once the model is installed.