Speak. Your words appear where the cursor is.
Voice dictation for Windows that runs entirely on your PC. Hold a key, speak in English or Portuguese, let go — and the text is typed into Outlook, Teams, VS Code or any other app.
- Free and open source
- No audio leaves your PC
- Works offline
Hold the key to “dictate” a sample sentence. Your real right Ctrl key works too.
Made for people who write all day
It works in every Windows app, because SpeechDesk types wherever the cursor is — no integrations needed.
English and Portuguese
The speech model covers 25 European languages, including English and European Portuguese, and detects on its own when you switch mid-sentence.
Spoken punctuation
Say “comma”, “full stop” or “new line” and you get the punctuation. The “um”s and “uh”s disappear.
Dictionary and snippets
Teach it names and technical terms. Say “my email” and the full address appears.
AI clean-up optional
When you correct yourself (“at 3, no, at 4”), only the final version stays. Fixes grammar and adapts the tone to the app.
Speak Portuguese, type English optional
Turn on translation in the menu and the text comes out in English, about one to two seconds after you let go of the key.
Private by default
Transcription, clean-up and translation run on your processor. History stays in a local file.
Shortcuts
A single key, and you can change it. AI clean-up and translation are switched on from the icon menu next to the clock.
| Shortcut | What it does |
|---|---|
| Hold Right Ctrl | Records while the key is held; types the text when you let go. |
| Double-tap Right Ctrl | Hands-free mode: records without holding the key. |
| Tap Right Ctrl | In hands-free mode, stops and types the text. |
| Esc | Cancels the recording. |
| Right Ctrl + another key | Cancels — it is a normal shortcut, like Ctrl+C. |
Models that run on your PC
Pick the engine in the settings. Models are downloaded once with a script.
Speech to text — default
NVIDIA Parakeet TDT 0.6B v3
- 25 European languages, including English
- Very fast on CPU
- ~0.7 GB · CC-BY-4.0 licence
Speech to text — alternative
OpenAI Whisper large-v3-turbo
- 99+ languages
- Built-in translation to English
- ~0.55 GB · MIT licence
- Fixes sentences Parakeet writes in the wrong language
Clean-up and translation — optional
Qwen 2.5 1.5B Instruct
- Runs inside the app, no server
- Turn it on or off from the menu
- ~1 GB · Apache 2.0 licence
Download
For Windows 10 and 11 (64-bit). No administrator rights needed. On first start SpeechDesk offers to download the speech model (about 0.7 GB) — once.
Installer
Adds SpeechDesk to the Start menu, with the option to start it with Windows.
Download installer (.exe)Portable
Unzip anywhere — a USB stick or a folder — and run SpeechDesk.exe. Settings and models stay next to the app.
Download portable (.zip)SpeechDesk is not code-signed yet, so Windows SmartScreen may show a warning the first time. Choose “More info” and then “Run anyway”. All versions are on the releases page.
Build from source
For developers. You need Windows 10 or 11 (x64), the .NET 8 SDK and 8 GB of RAM. Without AI, SpeechDesk uses about 1 GB; with AI, about 2.5 GB.
-
Get the code
git clone https://github.com/fmrgcarvalho/speechdesk.git cd speechdesk -
Download the speech model
Only once. Run it again with
-Engine aiif you want AI clean-up, or translation with Parakeet../scripts/download-model.ps1 -
Start it
An icon appears next to the clock. Open Notepad, hold Right Ctrl and speak.
dotnet run --project src/SpeechDesk.App
Project status
SpeechDesk is in active development (version 0.1). Early versions may have rough edges — feedback is welcome in the Issues.
Available
- Dictation with hold-to-talk and hands-free mode
- Parakeet and Whisper, running locally
- Rules: filler words, spoken punctuation, dictionary, snippets
- Optional AI clean-up and translation
- Local history
- Microphone choice and automatic gain for quiet microphones
- Language pinning: sentences in the wrong language are redone in yours
- Installer and portable version
Coming next
- Settings window
- Voice commands on selected text
- Automatic updates
- Text appearing while you speak