Whisperly for Windows

Hold a key, say what you mean, let go: the words appear wherever your cursor is. Speech becomes text on your own PC, through whisper.cpp. No account and no telemetry, and your recordings never leave your PC.

Public preview. Whisperly works, but it is still being finished: the installer is not code-signed yet, and Pro is not on sale.

Hold Right Ctrl, speak, let go. The text lands at your cursor.
Download for Windows
Checking for the latest build…
Whisperly in a minute and a half, narrated, with captions. Staged parts are labelled as illustrative in the film.

Install

  1. Open the installer. It is not code-signed yet, so SmartScreen says Windows protected your PC: choose More info, then Run anyway.
  2. It installs for your account only, into %LOCALAPPDATA%\Programs\Whisperly, with no administrator prompt. Whisperly starts when you sign in unless you untick that box.
  3. The setup guide takes about a minute: pick your microphone and watch the meter react, download a speech model, choose your key, and try it once before you rely on it.

Updating: Whisperly checks once a day and tells you when a new version is out. Run the new installer over the old one; your settings and models stay where they are. Uninstall from Windows Settings, under Apps, like any other app.

Using it

Hold Right Ctrl
Whisperly listens while the key is down; the capsule shows a live trace and a timer. Any Ctrl, Alt, Shift or Windows key can be the key instead.
Let go
The text is pasted at your cursor and your clipboard is put back. Settings can copy it instead, or type it character by character.
Esc
Cancels while recording or transcribing.
Ctrl+Alt+Shift+V
Search what you have dictated and paste it again. Pro.
Tray icon
Settings, history, and Transcribe File… for wav, mp3, m4a, aac, flac, aiff and wma recordings.

What it does

Whisperly's settings, on the Home page: ready to dictate, Right Ctrl as the key, and a readiness check for the microphone, the local engine and the active model.
Models
Seven whisper.cpp models, from Tiny (75 MB) to Large v3 Turbo (1.6 GB), and an experimental eighth trained for Danish. Each is checked against its SHA-256 before use.
Speed
Runs on your graphics card through Vulkan, and on the processor when there is no usable GPU. The model stays loaded between dictations.
Languages
100 languages and auto-detect with a multilingual model. Danish is handled end to end: punctuation by voice, decimal commas and names, on Large v3 Turbo — the model recommended for it, and free.
Dictionary
Replace what you say with what you mean, and keep a list of names and terms that should always be spelled your way.
Voice commands
Say “new line”, “period” or “open paren” and get the character.
Prompt mode
A second key that turns what you say into a cleaner prompt, shaped by a local model through Ollama.
History
Your last 200 dictations, kept on this PC. Turn it off or clear it in Settings.

Privacy

Your audio never leaves this PC, and there is no account or telemetry. Whisperly goes online only for these:

Update check
Once a day, one small request for the newest version number. Turn it off in Settings, under About.
Models
Downloaded from Hugging Face when you choose one.
Engine
If the bundled engine cannot start, you can install a processor-only build from GitHub.
Prompt mode
Local by default. If you switch its provider to a cloud model, the text you dictate in prompt mode is sent to that service, through your own sign-in.

Requirements

Windows
10 or 11, 64-bit
Disk
About 225 MB installed, plus 75 MB to 1.6 GB for each model
Graphics
Optional. A Vulkan-capable GPU makes transcription faster.
Microphone
Any input Windows can use. Whisperly follows your default device unless you pick one.

This build

Loading release notes…

Checksum

Verify the download in PowerShell with Get-FileHash WhisperlySetup.exe -Algorithm SHA256. The hash it prints must match this one.

—

A portable zip and every earlier version are on the releases page.