FreeDictate

Unlimited local speech-to-text dictation for Windows 11. Same push-to-talk feel as Wispr Flow, with no account, no cloud, and no weekly limit. Audio never leaves your PC.

Hold Ctrl+Win, speak, release. The text pastes at your cursor.

Install

Needs Windows 11 and Python 3.10 or newer from python.org (tick Add python.exe to PATH during setup). Not the Microsoft Store build.

One line. Open PowerShell and paste:

irm https://github.com/DavidHiFi/freedictate/releases/latest/download/Install.ps1 | iex

That downloads the release into %LOCALAPPDATA%\FreeDictate, installs the Python packages, fetches the Whisper model once (about 500 MB), and adds a single autostart entry.

Or by hand. Download freedictate.zip from the releases page, extract it anywhere, then:

powershell -NoProfile -ExecutionPolicy Bypass -File .\Install-FreeDictate.ps1 -Install

Add -Audit instead of -Install to check first without changing anything.

Use

1. Click any text field.
2. Hold Ctrl+Win and speak.
3. Release. Text pastes at the cursor and stays on the clipboard for a Ctrl+V fallback.
4. Esc cancels a take with nothing pasted.

Beeps mark the state: one beep when listening starts, a higher beep when text pastes, a low beep when a take is discarded or cancelled.

Start and stop

• Start: double click Start-FreeDictate.cmd, or log in again and the autostart entry launches it silently.
• Stop: Ctrl+C in its window, or end the pythonw process running freedictate.py in Task Manager.
• Takes append to transcripts.log in the install folder. Delete that file any time to clear history.

If you also use Wispr Flow

Quit Wispr Flow or rebind its push-to-talk key before starting FreeDictate. Both listen to Ctrl+Win, so running both pastes twice.

Configure

Edit config.json next to freedictate.py, then restart the app.

Key	Default	Meaning
model	small.en	Whisper model. tiny.en is faster, medium.en is more accurate.
device	cpu	Set cuda only if NVIDIA CUDA works on your PC.
compute_type	int8	Quantization. Leave it unless you know why.
language	en	Dictation language.
input_device	""	Mic name substring. Empty means the system default.
hotkey	ctrl+win	Fixed in this build.
auto_paste	true	false leaves text on the clipboard instead of pasting.
paste_delay	0.15	Seconds between copying and pasting.
beeps	true	Audible state cues.
min_seconds	0.5	Takes shorter than this are discarded.
silence_peak	0.005	Below this a take counts as silence and is discarded.
remove_fillers	true	Drops um, uh, you know, basically from transcripts.

FreeDictate reads your microphone. It never changes mic routing, audio apps, drivers, display settings, or services.

Commands

python freedictate.py                 hold Ctrl+Win to dictate
python freedictate.py --audit         read-only checks, changes nothing
python freedictate.py --list-devices  show input devices, changes nothing
python freedictate.py --mic-test 5    5 second level meter, changes nothing
python freedictate.py --preload       download and cache the model once

Uninstall

powershell -NoProfile -ExecutionPolicy Bypass -File .\Install-FreeDictate.ps1 -Uninstall

Then delete the install folder. No restart needed. pip uninstall faster-whisper sounddevice numpy pyperclip pynput clears the packages too.

How it works

pynput watches Ctrl+Win, sounddevice records while you hold it, faster-whisper transcribes locally, and pyperclip plus a Ctrl+V keystroke pastes the result. The model downloads from Hugging Face on first run and caches in your user profile. No server, no API key, no telemetry.

License

MIT