ON-DEVICE RECOGNITION
Your voice stays on your device.
Speech recognition runs on your computer. Download a model and leave cloud refinement and translation off to dictate offline.
YOUR VOICE. YOUR CHOICE.
Turn speech into text, on your own computer.
No subscription. Dictate offline after downloading a model. Connect your preferred AI only when you want refinement or translation.
Windows · macOS · Linux
A LITTLE ROOM TO THINK
Say what’s on your mind.
Give your ideas a place to land.
What will you write next?
Illustrative preview · Stay in the app you’re using
ON-DEVICE RECOGNITION
Speech recognition runs on your computer. Download a model and leave cloud refinement and translation off to dictate offline.
MIT LICENSED
Core dictation is free. The code is available under the MIT license to inspect, modify, and build yourself.
OPTIONAL AI REFINEMENT
Keep the transcription as it is, or connect your preferred provider for refinement and translation. Provider charges may apply.
MADE FOR YOUR EVERYDAY
Reply to a message, work through an idea, or describe what to build. Put your words in the app you’re already using.
Project chat
Let’s talk about the new design tomorrow morning. I’ll send my notes ahead of time.
A note to myself
A good tool leaves more room for the idea itself. Fewer interruptions, more space to think.
A prompt for AI
Make this page responsive: one column on mobile, two on desktop. Keep every button accessible by keyboard.
KNOW WHERE YOUR DATA GOES
Your audio is recognized on your computer. Sending text to a cloud AI is a separate choice that you make.
Explore privacy & dataNo audio upload for recognition · Local history
Receives text for refinement or translation
THE DETAILS THAT HELP
See live Chinese and English partial text in Streaming mode. Classic mode also supports Cantonese, Japanese, and Korean.
Add names, abbreviations, and specialist terms for streaming hotword bias and optional AI-assisted corrections.
Search, copy, and manage your dictation history locally. Records remain until you choose to delete them.
Voice Key is free and MIT licensed. Local recognition needs no ASR API key. Optional AI refinement and translation use your configured provider and may incur third-party charges.
Yes. Download a local model first and disable cloud refinement and translation. Core recognition and text insertion then work offline. Model downloads and update checks need a connection.
Classic mode supports Mandarin, English, Cantonese, Japanese, and Korean. Streaming mode supports Chinese and English with live partial text in the HUD. The interface is available in English and Chinese.
Local recognition does not upload microphone audio. If cloud refinement or translation is enabled, the relevant text and dictionary terms are sent to your configured provider. History stays on your device until you delete it.
macOS requires Microphone and Accessibility permissions. Current builds are unsigned and may trigger system warnings. Linux supports X11 and a dedicated Omarchy / Hyprland integration; other Wayland environments are not fully supported. See the getting-started guide.
MAKE ROOM FOR YOUR IDEAS
Download Voice Key for free. Choose your desktop platform.
Intel / Apple silicon
x64 · .exe
x86_64 · AppImage / pacman
Download a recognition model on first use. Current builds are unsigned; read the installation & permissions guide.