Voice typing that keeps up with you
Hold a key, speak, let go. The text appears in whatever window you are in, in about a tenth of a second, entirely on your own machine.
A small indicator appears while you dictate, and disappears when you are done. The wave follows your actual voice, so a flat line means the microphone is not hearing you.
Why it is fast
The model stays resident
Loading whisper-large-v3-turbo takes about 1.6 seconds. That happens once, in the background, never while you are waiting to type.
Nothing is uploaded
Transcription runs on your own GPU. No account, no API key, no network connection after the first launch. Your speech never leaves the machine.
It hears the start of words
A rolling pre-roll buffer captures the moment just before the key goes down, so a fast start is never clipped.
It knows when to say nothing
Low-confidence audio is discarded rather than guessed at, so silence and background noise do not turn into invented sentences in your document.
Controls
| Shortcut | What it does |
|---|---|
| Right Ctrl | Hold to dictate, release to type |
| Ctrl + Alt + Space | Toggle on and off, for long passages |
| Ctrl + Alt + Backspace | Delete what was just inserted |
| “new paragraph” | Spoken alone, inserts a line break |
| “scratch that” | Spoken alone, removes the last insertion |
Every shortcut is configurable, and spoken commands only act when the phrase is the whole sentence, so “add a new line to the file” is typed, not obeyed.
Installing
- Run the installer. It installs for your user only, so there is no administrator prompt.
- Windows may show a “Windows protected your PC” warning, because the app is not code-signed. Choose More info → Run anyway.
- On first launch it downloads the speech model, about 1.6 GB. The tray icon is amber while this happens, then turns slate when ready.
- Hold Right Ctrl and start talking.
What it needs
| Operating system | Windows 10 or 11, 64-bit |
| Graphics | NVIDIA GPU with ~2.5 GB free VRAM. Falls back to CPU automatically, considerably slower. |
| Disk | 1.4 GB for the app, plus 1.6 GB for the model |
| Microphone | Any. A headset or desk microphone gives noticeably better accuracy than a webcam. |
| Network | First launch only, to fetch the model |
Good to know
It will not type into programs running as administrator unless WhisperType is elevated too. That is a Windows security boundary, not something the app can work around.
The microphone is only held while you dictate. It is released about 30 seconds after you stop, so the Windows “in use” indicator goes away and other applications can have the device.
Right Ctrl is swallowed while the app is running, so it will not reach other programs. Change the shortcut in the settings file if you use it for something.