LesskeysWhisper on Windows: run it offline, step by step
Yes, Whisper runs offline on Windows with no setup: an app such as Lesskeys has a Whisper speech model built in, so you press a hotkey to dictate or add a file to transcribe. The audio never leaves the PC, and there is no account.
Windows 10 and 11 · No subscription
Last reviewed
Step by step
Run Whisper on your PC in four steps
- Install Lesskeys with Download for Windows. It needs Windows 10 (version 1903 or later) or Windows 11. A starter model is inside, so nothing else is needed.
- To dictate, click where the text should go, hold Alt+Space and talk, then release it. The text is pasted with Ctrl+V; in a window running as Administrator it stays on the clipboard for you to paste.
- To transcribe a recording, choose Transcribe and add the audio or video file. The first time there is a small extra download; after that it needs no internet.
- Open the transcript from Library. To check it ran on your PC, switch Wi-Fi off and repeat.
Installing on Windows? Step-by-step install guide.
Your PC
What runs it on a Windows PC
A graphics card with Vulkan support does the heavy lifting; without one the processor does it.
On a 64-bit Intel or AMD PC, Lesskeys uses a graphics card that supports Vulkan when it finds one, and falls back to the processor when it does not. The ARM64 build runs on the processor. Either way the model runs on this PC: transcribing without uploading explains why that matters.
- Offlineonce the model is on your PC
- No accountneeded
- Nothinguploaded
Two ways
The command line, or an app?
| What | whisper.cpp from the command line | Lesskeys |
|---|---|---|
| Setup | Build it yourself with CMake, then download a model | Run the installer; a starter model is inside it |
| Audio files | The command-line tool reads 16-bit WAV files; other files are converted first | Add an audio or video file as it is |
| Dictate into any app | No built-in way | Yes, with a hotkey |
| Transcribe a recorded call | A file at a time, after you record it | Records the call and transcribes it; no bot joins the call |
| Speaker labels | Not built in. An experimental option (tinydiarize) needs a special English model and marks where the speaker changes, not who spoke | Yes: you set how many people spoke, and each voice gets its own label |
Whisper writes only the words; speaker labels come from a separate step, covered in speaker diarization.
Tell the people you record; some places require everyone's consent.
Sources, read on 21 October 2026
- whisper.cpp, the open-source project (Windows among the supported platforms, the CMake build, 16-bit WAV input, Vulkan needs a driver, speaker segmentation via tinydiarize (experimental))
Models
Which model, how big?
Tiny is built in, so dictation works before any download. Bigger models are one download away.
Lesskeys offers Whisper models from Tiny (about 75 MB) up to Large V3. Setup suggests Large V3 Turbo Q8, about 834 MB. Which one is most accurate was measured on a Mac, not on Windows: see Which Whisper model to run on a Mac.
Questions
Questions about Whisper on Windows
Does Whisper run on a Windows PC without a graphics card?
Yes. Lesskeys uses a graphics card that supports Vulkan when it finds one and otherwise runs the model on the processor, so a PC without a suitable graphics card still works.
Does Whisper work offline on Windows?
Yes. Once the model is downloaded, Whisper transcribes entirely on your PC with no internet connection, and your audio never leaves the device.
Do I need to build whisper.cpp to use Whisper on Windows?
No. Building it is the technical route, with CMake and a model download. An app such as Lesskeys runs Whisper for you and adds dictation, call transcription and speaker labels.
Is Lesskeys the same as Whisper on Windows?
No. Lesskeys runs Whisper on your PC and wraps it with system-wide dictation, call and video transcription and speaker labels, all offline. It is an app around the models, not the models themselves.
More