Short answer: For voice-to-text on a Mac where audio never leaves the machine, you have three options: built-in macOS Dictation (on-device on Apple Silicon, fine for short bursts), a cloud dictation app (most accurate, but your audio is uploaded), or a dedicated on-device app that runs local models with the ergonomics you’d actually use all day. If privacy and daily speed both matter, an on-device app is the sweet spot — which is exactly why I built one for myself.
When the built-in option wasn’t enough and I didn’t want my voice going to someone’s server, I built my own — TobiVoice, a native macOS app that transcribes on local models. It’s a personal build I use daily, not a product you can download today. But the reason it exists is the useful part here, so let me compare the honest options.
The three options
| Option | Where it runs | Privacy | Best for |
|---|---|---|---|
| macOS Dictation | On-device (Apple Silicon) | Audio stays local | Occasional short dictation |
| Cloud dictation apps | Cloud servers | Audio uploaded | Maximum accuracy if you accept the cloud |
| Dedicated on-device app | On-device (local models) | Audio never leaves your Mac | Fast, private, all-day dictation |
Built-in macOS Dictation
It’s already there, it’s free, and on Apple Silicon it runs on-device for many languages. For firing off a quick sentence, it’s perfectly good.
Where it falls short is sustained, everyday use: accuracy drifts on longer passages, there’s no push-to-talk workflow to lean on, and you can’t teach it a personal dictionary or have it adapt to the app you’re writing in.
Cloud dictation apps
The large cloud models are accurate — often the most accurate option. The trade-off is right there in the name: your audio is uploaded to someone else’s servers, and most of these tools are subscriptions.
If your work is sensitive — client calls, health, legal, anything under confidentiality — “we send your voice to the cloud” is the line you may not want to cross.
A dedicated on-device app
This is the gap I built TobiVoice to fill for myself: a native macOS app (macOS 14 or later) that transcribes on local AI models, so processing is fully offline and audio never leaves the Mac. What makes an on-device app a daily driver rather than a one-off:
- Near-instant transcription on local models — no cloud round-trip.
- Global shortcuts for quick recording and push-to-talk.
- Context-aware modes that adapt to the app you’re writing in.
- A personal dictionary for custom terms and replacements.
- 100% offline — nothing is uploaded, ever.
It’s the combination that matters: cloud-app ergonomics (shortcuts, speed, custom terms) without the cloud. None of that needs a server — it just needs someone willing to build it properly on-device instead of reaching for the easy cloud API.
How to choose
- Just need the odd sentence? Built-in Dictation is fine — you already have it.
- Chasing maximum accuracy and don’t mind the cloud? A cloud app will edge ahead.
- Want speed and privacy for all-day use? A dedicated on-device app is the fit — the reason I built my own rather than settle for either end.
Same principle runs through everything I build: the useful part should live on your device, built by someone who knows how to do it there. If you’re on Android, TobiBoard vs Gboard makes the same case for your keyboard.