Ctrl Dictate

Hold Ctrl, talk, let go. Your words are typed wherever your cursor is, with the ums, uhs, and false starts already cleaned out.

Download for Windows

Free · Windows 10/11 · 6 KB zip

You say

~~Um, so~~ I think we should, ~~uh,~~ meet on ~~Tuesday, no,~~ Wednesday to go over ~~the~~ the numbers.

It types

I think we should meet on Wednesday to go over the numbers.

Works everywhere you type

Email, Word, Google Docs, Slack, Teams, text boxes on any website. If you can put a cursor in it, you can dictate into it. Your normal Ctrl shortcuts still work: pressing another key while holding Ctrl (like Ctrl+C) never triggers dictation.

Setup

Install Python (free) from python.org. On the first installer screen, tick “Add python.exe to PATH”.
Get an Anthropic API key at console.anthropic.com (Settings → API Keys). This powers the clean-up step. You pay Anthropic directly, roughly a cent or less per dictation.
Unzip the download somewhere permanent, like your Documents folder, and double-click Start Dictation.bat.
The first run installs everything and downloads the speech model (a few minutes, about 500 MB), then asks for your API key. When the window says Ready, you're set. Leave it open or minimized while you work.

Privacy

Your voice is turned into text on your own computer. Audio never leaves it.
Only the text is sent to Anthropic's Claude for clean-up.
Your API key is saved only on your computer, in the app folder.
Nothing is sent to me or this website.

Questions

Can I use it without an API key?

Yes. Press Enter when it asks for a key and it will type exactly what you said, without the AI clean-up.

It won't type into one particular program.

That program is probably running as administrator. Right-click Start Dictation.bat and choose “Run as administrator”.

Nothing happens when I hold Ctrl.

Make sure the Ctrl Dictate window says “Ready”, and that Windows lets apps use your microphone (Settings → Privacy → Microphone).

I entered the wrong API key.

Delete api_key.txt in the app folder and start it again. It will ask for a new one.

Can I make it faster or more accurate?

Open dictate.py in Notepad and change WHISPER_MODEL near the top to "base.en" for faster, or "medium.en" for more accurate.