Open Dictation records from a chosen microphone, transcribes the finished recording, and pastes the result at the cursor. Its notch panel shows recording and transcription state on supported MacBooks. Macs and displays without a notch use audio feedback.
- Whisper runs locally and remains the default speech engine.
- Parakeet TDT v2 handles English, while Parakeet TDT v3 supports 25 European languages. Both are optional and run locally through FluidAudio 0.15.6.
- Parakeet models download only after you choose Download and Prepare Model. Parakeet prewarming uses installed models without downloading. Whisper retains its automatic model selection and upgrade behavior.
- The global shortcut supports press-to-start/press-to-stop and hold-to-talk modes.
- Microphone choices use stable Core Audio device identifiers. If a selected device is missing, recording stops with a clear error instead of changing inputs.
- A personal dictionary supplies recognition hints to Whisper and compatible cloud services. Literal replacements work with every engine, including Parakeet.
- Transcript cleanup is deliberately narrow. It removes known speech-to-text control and noise markers without broad punctuation rewrites. Optional filler-word removal is off by default and avoids rewriting fragments inside other words.
- Cloud presets cover OpenAI GPT Transcribe, GPT-4o mini Transcribe, OpenAI Whisper, and Groq Whisper Turbo. You can also set an OpenAI-compatible base URL and model name.
- If Accessibility access is available, Open Dictation pastes at the cursor and restores the previous clipboard contents. Without it, the transcript stays on the clipboard.
- macOS 14 Sonoma or later
- Apple silicon Mac
- Microphone permission
- Accessibility permission for automatic paste
-
Drag Open Dictation to Applications.
-
For an unsigned build, remove its quarantine attribute:
xattr -rd com.apple.quarantine /Applications/OpenDictation.app
-
Launch the app and grant the requested permissions.
The default shortcut is Option + Space. In toggle mode, press once to record and again to transcribe. In hold mode, keep the shortcut down while speaking and release it to transcribe. Press Escape to cancel an active dictation.
Open Settings to choose a microphone, shortcut, activation mode, language, local engine, or cloud service. Whisper models are stored under:
~/Library/Application Support/com.opendictation/Models/
See dictation validation for the manual acceptance and benchmark protocol. Third-party code and model attribution is listed in third-party notices.