Medical Dictation Software for PC,
Free & 100% Offline.

Everything runs on your machine. The speech model is a local process, the recording is written to a temporary file on the same disk and deleted once the text is produced, and the result is typed directly into the field you are working in rather than copied through the clipboard, so it is not exposed to clipboard managers or to whatever is being screen-shared. There is no account, no upload, and no network request during transcription. If you want to prove it rather than take our word for it: disconnect the machine from the network and dictate. It behaves identically.

Yes. Whisper Dictator sends real keystrokes to whichever field currently has focus, so it works anywhere you can type on Windows: Epic, Cerner, Meditech, Athena, eClinicalWorks, browser-based portals, Word, or your own templates. There is no per-EMR integration to buy, configure, or renew, and nothing to install inside the EMR itself. Because it types rather than pastes, it also works in fields that deliberately block paste, and it leaves your clipboard untouched.

It is built on OpenAI’s Whisper, which was trained on a very broad range of speech and handles common drug names, anatomy, and everyday clinical abbreviations well. Being straight with you: it is a general-purpose model, not one tuned to a single specialty’s vocabulary, so an unusual brand name or a niche acronym can come out wrong. In practice that makes it well suited to narrative documentation you review before signing, and less suited to unattended dictation you never read. The Large model is the most accurate for dense terminology, and an NVIDIA GPU makes it fast enough to keep up with live speech.

Both ways. You can dictate live while the patient is still in the room, which is where most of the time is actually saved, because the note is finished before you leave rather than after clinic. You can also point it at a consultation you already recorded and transcribe the entire file locally. In both cases the audio is processed on the machine that holds it and the temporary file is removed afterwards.

The Lifetime license covers up to 3 workstations and the Annual covers 2, which is enough for a small practice or for one clinician moving between a desk, an exam room, and a laptop. Beyond that, each provider gets their own license. There is no server to stand up and no central console to administer, because there is no shared component: each install is self-contained. Get in touch for volume pricing.

The installer is a standard Windows package and needs no special privileges beyond a normal install. The only permission the application requests is microphone access; it does not need inbound network access, a service account, or an exception in your EMR. Paid licenses include priority support, and we can supply documentation for a desktop team that wants to review it before rollout.

Dragon Medical is sold as a per-clinician subscription and is the more specialised product, with a vocabulary tuned to clinical speech and deeper EMR-specific integrations. Whisper Dictator is deliberately a different trade: $39.99 a year or $99 once, for every feature, on 2 or 3 workstations, with a free tier of 7 transcriptions a day that does not expire and needs no card. If your documentation is narrative text that you review before signing, the cheaper tool is very likely enough. If you depend on specialty vocabulary tuning or on commands that drive the EMR itself, that is genuinely what the expensive one is for.

The recording is written to a temporary file on the same machine, read by the local speech model, and then deleted. Nothing is uploaded, nothing is archived, and no copy is kept on our side, because no copy ever reaches our side: we operate no server that receives audio or text. If your own policy requires you to retain a recording of an encounter, you have to record and store it yourself deliberately, because the software will not quietly keep one for you.