Five questions, five honest answers.
Most dictation software sells a feeling. These are the five things people actually ask before they install anything, each with its own page and its own evidence. Pick the one that sounds like your hesitation.
Will my source code end up in somebody else’s model?
No API key exists, because there is no API to key into.
Transcription runs on your CPU and the AI rewriting runs on a model on your own GPU through Ollama. Pull the network cable and it behaves identically, which is the only version of that claim you can check yourself in about a minute. Commit messages, PR descriptions, tickets and docs, from the version you would have typed badly at 1am.
Read the full answerOffline dictation for WindowsDoes it still work with no internet?
After the one-time model download it needs no network at all.
Not a degraded offline mode - the same product. Nothing is streamed anywhere for processing, so a plane, a locked-down box, a hospital network or a lab with no internet all behave the same as your desk. Built for the people who cannot send audio out and are tired of being told to trust a privacy policy about it.
Read the full answerGerman, French and Spanish dictationCan I dictate in German, French or Spanish offline?
All three are on Parakeet’s list of 25 languages, and Parakeet runs on your CPU.
Parakeet, the default engine, works out which of its 25 languages you are speaking, and Whisper keeps each recording in the language you choose. Both run on your own PC, so German, French and Spanish dictation keeps working with the Wi-Fi off once setup is done. We have not measured accuracy in these languages, so try your own voice first.
Read the full answerDictation without a GPUDo I need a graphics card for this?
Not for dictation. 11.9% word error on the published clips, on CPU cores.
NVIDIA’s Parakeet v3 runs on any modern CPU and uses no VRAM at all, so the no-GPU path is the whole product rather than a trial mode. It also means the card you do have stays free for the build, the container, the training run or the game. A GPU is what unlocks the AI rewriting, not what unlocks typing.
Read the full answerDictation without a subscriptionWhat does this cost me in three years?
$49 once, against $432 to $540 for three years of Wispr Flow Pro.
The only argument on this site that survives complete distrust, because you can check it by dividing. There is no renewal, no per-minute meter, and no account needed for the app to work. We can charge once because the expensive part runs on your hardware rather than ours - so there is no running cost to pass on.
Read the full answerPrivate, offline dictation on Windows 11: what works in 2026
The free routes first: voice access, Win+H voice typing and open-source apps, each checked against its own page, then where Voxmelt fits and what it costs.
Read the guideDictation for writers on Windows: long-form drafts without the cloud
For novelists, bloggers and students: voice access, Win+H and Word’s Dictate first, then Dragon, Wispr Flow and Superwhisper, then Voxmelt, with punctuation, listening time and a long-draft workflow.
Read the guideLocal AI dictation on Windows: speech and cleanup, both on your PC
Fluid dictation on Copilot+ PCs, OpenWhispr and a do-it-yourself whisper.cpp and Ollama setup first, then Voxmelt, plus how to check that the AI step, not just the speech, stays on your PC.
Read the guideOne machine. One payment. No meter.
Whichever of those five brought you here, it is the same app: dictation on your CPU, optional AI polish on your own GPU, nothing leaving the machine, and a price you pay once. Windows 10 and 11, free tier forever, 15-day full-Studio trial with no card.
Direct installer, for PCs without the Store; Windows will warn because it is not code-signed.