Voice Dictation
Speaking is faster than typing. Now you can dictate directly into RiteMark — and unlike other tools, your voice never leaves your computer.
RiteMark's voice dictation runs entirely on your Mac using whisper.cpp. No cloud, no API calls, no audio data sent anywhere. Ever.
Enable Voice Dictation
Voice dictation is an experimental feature, so you'll need to flip it on first.
Open Settings (gear icon in the sidebar, or Cmd+,), find the Voice Dictation section, and toggle it on. Once enabled, a microphone button appears in your editor toolbar.
💡 Tip: "Experimental" just means we're still refining it based on feedback. It works well — we use it ourselves every day.
How to Dictate
Click the microphone button in the toolbar and start speaking. Text appears in real-time as you talk. Click the button again to stop.
On first use, RiteMark downloads the speech model (~244MB one-time download). After that, everything runs offline — no internet needed.
Your dictated text goes wherever your cursor is. You can freely mix typing and dictation, switching between them as naturally as you'd switch between keyboard and mouse.
🎉 That's it! You're dictating privately on your own machine. No accounts, no API keys, no cloud.
Changing Language
RiteMark supports Estonian and 50+ other languages for dictation.
Click the dropdown arrow (chevron) next to the mic button to see your recent languages, or click "More languages..." for the full list. RiteMark remembers your last used language, so you won't need to pick it again next time.
Estonian is the default and works well out of the box. The model handles multiple languages without needing additional downloads.
Dictation Settings
Click the dropdown arrow next to the mic button and select "Settings..." to fine-tune your experience.
Chunk duration controls how often audio is processed. Think of it as a responsiveness slider:
- 3 seconds — Most responsive, text appears very quickly
- 5 seconds — Balanced (this is the default)
- 10 seconds — Most accurate, better for complex or technical speech
Shorter chunks feel snappier. Longer chunks give the model more context to work with. Try different settings — you'll quickly find what matches your speaking style.
Privacy and Security
Here's exactly what happens with your audio — no fine print, no surprises:
- Audio is captured from your microphone and held in memory
- The whisper.cpp model processes it on your Mac's hardware
- Transcribed text is inserted into your document
- Audio data is discarded — never written to disk, never transmitted
No internet connection is used during dictation (only for the initial model download). No account or API key is needed. Your voice stays yours.
Troubleshooting
"Microphone Access Required" dialog appears
macOS needs your permission before any app can use the microphone — that's a good thing.
Click "Open System Settings" in the dialog, navigate to Privacy & Security > Microphone, and enable the toggle for RiteMark. You'll need to restart RiteMark after granting permission.
Model download stuck or failed
The speech model (~244MB) downloads on first use. If it gets stuck, check your internet connection, then close and reopen RiteMark to retry. The model is stored at ~/.ritemark/models/ — you can delete this folder to force a fresh download if needed.
Dictation not starting after clicking mic
Make sure the speech model finished downloading (check Settings > Voice Dictation for model status). Verify microphone permissions are granted in System Settings. If both look good, try restarting RiteMark.
Poor transcription quality
A few things that help: speak clearly at a normal pace, reduce background noise, and try a longer chunk duration (10s) for better accuracy. Speaking in complete sentences also helps — the model uses surrounding context for better results.
Estonian transcription issues
Confirm you've selected "Estonian" from the language dropdown (not a similar language). The model handles standard Estonian well, though heavy dialect may reduce accuracy. Speaking in full sentences gives the best results.
Quick Reference
| What you want to do | How to do it |
|---|---|
| Start dictating | Click mic button in toolbar |
| Stop dictating | Click mic button again |
| Switch language | Dropdown arrow next to mic → select language |
| Adjust responsiveness | Dropdown arrow → Settings → Chunk duration |
| Check model status | Settings > Voice Dictation |
| Remove speech model | Delete ~/.ritemark/models/ folder |
Technical Details
| Detail | Value |
|---|---|
| Speech model | OpenAI Whisper large-v3-turbo (local) |
| Runtime | whisper.cpp (native Apple Silicon) |
| Audio format | 16kHz mono PCM, processed in memory |
| Model storage | ~/.ritemark/models/ (~1.5GB on disk) |
| Model download | ~244MB (one-time) |
| Languages | 50+ including Estonian |
| Internet required | Only for initial model download |
| Audio data sent externally | Never |
Related
- Export — Export options including the new Copy as Markdown
- Core Editor — Basic editing features
- Keyboard Shortcuts — Speed up your workflow
Happy writing!