Project · In development
Parley
Speech-to-text dictation that runs on your own computer, so your words are never harvested. Hold Control + Option, say what you want to write, and let go. Your words appear where you were typing. It runs smoothly on a Mac, with no account and no network.
Say it. Keep writing.
Speak where you type
A small sound meter at the bottom of the screen shows Parley is hearing you. If nothing that takes text is focused, a small window shows your words with a Copy button.
A second listen at every pause
Three speech models listen together. One hears you live; at each pause, two more re-hear the same stretch and vote on the words.
Leave out the hesitations
A set of rules tidies the text as you talk, dropping fillers like “um”, stutters and repeated words. Say “no wait” to correct yourself.
Punctuation in your own voice
Say “comma” or “new line” to shape your text, or “scratch that” to remove what you just said. Parley also handles words spelled out letter by letter.
Times and prices, written out
Speak numbers, times and prices naturally. “Three thirty pm” becomes 3:30 PM.
Teach it your terms
Add names and terms Parley should know. You can change the hotkey to a shortcut that suits you.
Your words stay on your Mac.
No account, no subscription, and no network code at all. Every model runs on your Mac, so what you say never leaves your computer.
Written in Rust, for macOS.
Built with these projects.
Parley ships these models and libraries, all unmodified, each under its own license:
- Kroko ASR English streaming model by Banafo — CC BY-SA 4.0.
- NVIDIA Parakeet TDT-CTC 110M — CC BY 4.0.
- Moonshine v2 tiny by Moonshine AI / Useful Sensors — MIT.
- sherpa-onnx by k2-fsa — Apache 2.0, with ONNX Runtime (MIT).
- Word frequencies counted from Tatoeba (CC BY 2.0 FR) and WikiText-103 (from Wikipedia, CC BY-SA 3.0).
- English word list from SCOWL by Kevin Atkinson.