Dictation

Dictation is a small Windows app that turns a global hotkey and your voice into text in the application that has focus. Speech recognition runs locally on the device; audio is not sent to a cloud service.

Use it

Requirements:

• 64-bit Windows 11
• A working microphone
• .NET 10 SDK for building and running from source

The default shortcut is Ctrl + Alt + Space. In Hold mode, press and hold the chord while speaking, then release it; the transcript is inserted at the focused caret. Toggle mode starts recording on the first press and transcribes on the next press. The shortcut, microphone, language, and appearance can be changed in Settings.

Settings and transcript history stay in the user's local %LocalAppData%\Dictation folder. Dictation remains unelevated, so Windows will not allow it to inject text into an elevated administrator window.

Local speech recognition

Dictation uses sherpa-onnx 1.13.5 with CPU inference:

• English uses the int8 NVIDIA Parakeet TDT 0.6B v3 model.
• Auto, Dutch, and Arabic use an int8 Whisper Small ONNX conversion.

The models are downloaded separately by scripts/install-model.ps1; they are not included in this repository. Their terms are documented in THIRD-PARTY-NOTICES.md.

Build, test, and run

From PowerShell:

.\scripts\install-model.ps1
dotnet restore .\tests\Dictation.Tests\Dictation.Tests.csproj --configfile .\NuGet.Config --ignore-failed-sources -p:NuGetAudit=false
dotnet build .\Dictation.slnx -c Release --no-restore -m:1 -p:NuGetAudit=false
dotnet test .\tests\Dictation.Tests\Dictation.Tests.csproj -c Release --no-restore --no-build -m:1 -p:NuGetAudit=false
dotnet run --project .\src\Dictation\Dictation.csproj -c Release --no-restore

To create a standalone folder, publish a Release build and install the models into that folder:

dotnet publish .\src\Dictation\Dictation.csproj -c Release -r win-x64 --self-contained false -o .\publish
.\scripts\install-model.ps1 -Destination .\publish\models
.\scripts\register-start-menu.ps1 -Source .\publish

The project is released under the MIT License. See THIRD-PARTY-NOTICES.md for dependency and model attribution.