Skip to content

Whisper Voice

Local speech-to-text for Windows using OpenAI Whisper — like Wispr Flow, but free and fully offline.

Windows .NET License

Features

  • Push-to-Talk & Toggle Mode — customizable global hotkey (default: Win+D)
  • 100% Local — no cloud, no API keys, complete privacy
  • Single Process — one .NET executable, no Python, no FFmpeg, no Node.js
  • GPU Acceleration — CUDA support for NVIDIA GPUs with automatic CPU fallback
  • Auto-Paste — transcribed text is copied to clipboard and pasted at cursor
  • Multiple Languages — German, English, and many more, or auto-detect
  • System Tray — runs quietly in the background
  • Floating Status Bar — always-visible recording indicator
  • Auto-Updates — delta updates via Velopack

Requirements

  • Windows 10/11 (64-bit)
  • NVIDIA GPU with CUDA (optional, recommended for faster transcription)

Installation

Download the latest installer from Releases.

The installer handles everything — no dependencies to install manually.

Build from Source

git clone https://github.com/StaniB88/WhisperVoice.git
cd WhisperVoice
dotnet run --project src/WhisperVoice

Publish (self-contained)

dotnet publish src/WhisperVoice -c Release -r win-x64 --self-contained -o publish

Usage

  1. Launch — the app opens with a setup wizard on first run
  2. Hold Hotkey — press and hold your configured hotkey (default: Win+D)
  3. Speak — your audio is recorded locally
  4. Release — audio is transcribed using Whisper
  5. Done — text is pasted at your cursor position

Configuration

Settings are stored in %APPDATA%/whisper-voice/config.json

Option Description
Whisper Model tiny (fast) → large (best quality)
Language German, English, auto-detect, and more
Hotkey Any key combo, recorded in settings
Toggle Mode Press once to start, press again to stop
Auto-Paste Automatically paste with Ctrl+V after transcription
Floating Bar Show/hide the floating status indicator
Theme Multiple dark themes to choose from

Whisper Models

Models are downloaded on first use and stored in %APPDATA%/whisper-voice/models/.

Model Size VRAM Speed Quality
Tiny ~75 MB ~1 GB ~10x Basic
Base ~142 MB ~1 GB ~7x Good
Small ~466 MB ~2 GB ~4x Better
Medium ~1.5 GB ~5 GB ~2x Great
Large v3 ~2.9 GB ~10 GB 1x Best
Large Q5_0 ~1.1 GB ~5 GB ~1.5x Great
Large Turbo ~1.5 GB ~6 GB ~8x Best
Large Turbo Q5_0 ~574 MB ~3 GB ~10x Great

English-only variants (Tiny, Base, Small, Medium) are also available for improved accuracy on English speech.

Architecture

Single-process .NET 10 application — no child processes, no IPC, no HTTP servers.

User presses hotkey
    │
    ▼
Win32 SetWindowsHookEx (global keyboard hook)
    │
    ▼
NAudio (16kHz/16-bit/mono WAV capture)
    │
    ▼
Whisper.net (CUDA → CPU fallback)
    │
    ▼
Clipboard + SendInput (auto-paste at cursor)
Component Technology
UI Avalonia UI 12
MVVM CommunityToolkit.Mvvm
Transcription Whisper.net (whisper.cpp bindings)
Audio NAudio
Hotkeys Win32 P/Invoke
Updates Velopack

Troubleshooting

Slow transcription

  • Use a smaller model (Tiny or Base)
  • Install NVIDIA CUDA drivers for GPU acceleration
  • Ensure GPU drivers are up to date

Hotkey not working

  • Some key combinations are reserved by Windows
  • Try a different hotkey in Settings → Hotkey → Record

Model download fails

  • Check your internet connection
  • Models are downloaded from Hugging Face — ensure it's not blocked

License

Copyright (C) 2026 AnyAutomation.

Whisper Voice is licensed under the GNU Lesser General Public License v3.0 or later (LGPL-3.0-or-later). See LICENSE / COPYING.LESSER and COPYING for the full license texts.

Third-party components are listed in THIRD-PARTY-NOTICES.md.

Credits

Support

If you find Whisper Voice useful, consider buying me a coffee.

About

Lokale Spracherkennung mit OpenAI Whisper

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Releases

Sponsor this project

Packages

Contributors

Languages