Most dictation apps work the same way. You talk, your audio goes up to a cloud server, and text comes back. That's fine for a shopping list. It's less fine when you're dictating a legal draft, patient notes, or an email about a deal that hasn't been announced.
So I built Quiet Dictation: hold a hotkey, speak, release, and clean, punctuated text appears wherever your cursor is. The difference is that speech recognition runs entirely on your own laptop. No audio is uploaded or stored anywhere else.
Why on-device matters
- Privacy: your voice never leaves your machine, so there's nothing on a server to leak, subpoena or train on.
- Offline: it works on a plane, behind a hospital firewall, or on terrible cafe Wi-Fi.
- Cost: with no servers transcribing audio, it can be cheap. It's $5/month or $39/year. Wispr Flow is $15/month.
What it does
- Works in any app: Word, Outlook, Gmail, Google Docs, Slack, WhatsApp, your browser
- Double-tap the key for hands-free mode when you're dictating something long
- Personal dictionary for names and terms it should always get right
- Meeting recording and transcription
- Tuned for non-American and non-native English accents. I'm in South Africa, and most dictation tools treat my accent as an edge case.
It runs on Mac (Apple Silicon, macOS 13+) and Windows 10/11. There's a 14-day free trial, no card needed.
If you type all day, try it and tell me what breaks: quietdictation.com
Top comments (1)
The on-device boundary is clearest when the product can expose a small, inspectable privacy receipt after each session: selected recognition mode, network activity attributable to transcription, retained local artifacts, and the destination application. That would let someone validate the claim on their own machine, including during a fallback or meeting-recording workflow.