Parakeet speech to text on Windows
NVIDIA Parakeet TDT v3, running on your own PC.
Parakeet TDT v3 is a speech recognition model from NVIDIA. VinPUT can run it locally: a 669 MB download that knows 25 European languages and works with no internet connection and no account. It does not need an NVIDIA card.
Where it runs
Parakeet uses the same engine as Whisper, whisper.cpp. On a PC without a suitable graphics card it runs on the processor. If the PC has a card, Settings, Models offers Run on the graphics card, ticked by default, which downloads an 18 MB Vulkan build that works on NVIDIA, AMD and Intel cards.
I do not publish speed figures, because they depend on your machine. The Models page rates Parakeet for your PC as running well, maybe slow or too heavy, and the bottom left of Settings shows whether it is on the processor or the card.
Languages
You do not choose a language for Parakeet. It detects the language itself, and it knows 25 European languages. If one of the languages you list under Settings, General, Languages is missing from its list, the Models row says Does not know and names it. For a language outside those 25, use Whisper large-v3 turbo, which knows 99.
When the words appear
Parakeet recognizes the whole dictation when you stop, then types it. Whisper types at each pause. That suits Parakeet to short thoughts, a sentence or a message, where you want the finished text in one go. VinPUT starts the Parakeet program for each dictation.
The dictionary, snippets, spoken formatting and cleanup all apply to Parakeet's text, as they do to every model.
Start
Open Settings, Models, find NVIDIA Parakeet TDT v3 under On this PC, click Download, and then Use when it says Downloaded. Tap F1 in any text field and speak. The first run also fetches the engine. See the documentation for the full list of models.
Questions
Does Parakeet need an NVIDIA card?
No. It runs on the processor, or on an NVIDIA, AMD or Intel card through the Vulkan build of the engine.
Which languages does it know?
Twenty-five European languages. It detects the language itself. The Models page tells you if it does not know one of the languages you set.
Does it type while I speak?
No. Parakeet types the whole dictation when you stop. Whisper is the one that types at each pause.
Should I choose Parakeet or Whisper?
Parakeet if your language is among its 25 and you like to get the text in one go. Whisper if you need another language, or want text while you are still speaking.
Related pages
- Whisper dictation The Whisper sizes VinPUT lists and how they run.
- Offline speech to text What works with no connection, and what needs one.
- Superwhisper for Windows, compared Two dictation apps with local models, side by side.
- Voice typing on Windows 11 and Win+H What the built-in option does, and when it is enough.