Dictation for Windows
talkflow is a free, open-source dictation app for Windows. Hold a shortcut, speak, and your words are typed into whatever app you're using. The speech recognition runs on your own PC.
What it does
Hold Ctrl + Win, say what you want to write, and let go. talkflow types your words into the focused text field: a chat, an email, a document, a code editor, a browser. You can pick a different shortcut in Settings. There is no account, no trial and no word limit.
What you need
- Windows 10 (version 1809) or later, or Windows 11.
- Any PC that runs those. There is one installer, for x64, and Arm-based Windows PCs run it through Windows' built-in emulation.
- A microphone, and about 490 MB of disk space for the English speech model, plus about 575 MB if the optional, more accurate final-pass model is downloaded.
The installer puts talkflow in your own user folder, so it doesn't need administrator rights.
How it works on your PC
The installer includes whisper.cpp's speech engine. talkflow runs it in the background, listening only at 127.0.0.1 (your own computer), and loads OpenAI's open Whisper model into it. On first run it downloads the English model, small.en, once. Your audio goes from the microphone to that engine and never over the internet.
talkflow measures how fast your PC is and tunes itself to match. It can use your graphics card through Vulkan when the driver supports it, and falls back to the processor. On a fast enough PC it also uses a larger model, large-v3-turbo, for the final text; on a slower PC it skips that model and turns off the live caption so the final text isn't delayed. More detail is in how talkflow works.
Microphone access
Windows doesn't show a permission prompt for desktop apps. Instead, Settings > Privacy & security > Microphone has switches that either allow them or silently give them silence. talkflow checks those switches, and if one is off, it tells you which and opens the right Settings page.
How the text is typed
Short dictations are typed as keystrokes. Longer ones are pasted in one go, because Windows apps can garble long runs of synthetic keystrokes: talkflow saves your clipboard, pastes, then puts your clipboard back.
What leaves your PC
Not your voice, and not your words. The app makes only these requests:
- A one-time install count from official releases: one empty request with no ID and nothing about you or your PC, sent once and never again.
- Update checks with GitHub. Updates install only when you choose to.
- Model downloads from Hugging Face during setup.
- Only if you add your own key: OpenAI for transcription, or Anthropic for punctuation. Off unless you add one.
See the privacy policy and the install counter's public documentation.
Good to know
- It is built for English; its main model, small.en, is English-only.
- It types into the focused text field of most apps, but not every app has been tested.
- There is also a Mac version: see offline dictation for Mac.
