System-wide voice-to-text dictation for Windows using AI

Enjoying VocalFold? Support the development of this free, open-source project!

☕ Buy me a coffee on Ko-fi — Your support helps keep this project alive and growing!

VocalFold is a Windows desktop application that transcribes your voice to text using Whisper.NET AI. Press a global hotkey, speak, and your words appear as text at the cursor position in any application.

• 🎤 Voice Recording: Activate with global hotkey (Ctrl+Windows by default)
• 🤖 AI Transcription: Local processing using Whisper.NET (no cloud services)
• ⚡ GPU Acceleration: Support for NVIDIA (CUDA), AMD (Vulkan), and Intel (Vulkan) GPUs
• ⌨️ Text Output: Types text at cursor position (works in any app)
• 🔒 Privacy-First: All processing happens locally, no data leaves your machine
• 🎯 Background Operation: Runs in system tray, always available
• 🚀 Windows Startup: Optional auto-start with Windows
• 🎨 Modern Web UI: Configure settings via beautiful web interface
• 📝 Keyword Replacement: Create shortcuts for frequently used phrases
• 🗂️ Category Organization: Organize keywords into collapsible categories

• Windows 11 (recommended)
• Windows 10 (supported)

VocalFold automatically selects the best available GPU runtime:

NVIDIA GPUs (CUDA):

• Supported: RTX 20 series or newer (RTX 2060, 3060, 3080, 4080, etc.)
• Requires: NVIDIA CUDA Toolkit 12.x (Download)
• Performance: Excellent (~0.5s for 5s speech with Base model)
• VRAM: 4GB minimum, 8GB+ recommended

AMD GPUs (Vulkan):

• Supported: Radeon RX 6000 series or newer (RX 6700 XT, 6800 XT, 7900 XTX, etc.)
• Requires: Latest AMD Adrenalin drivers with Vulkan support (Download)
• Performance: Good (~1-2s for 5s speech with Base model)
• VRAM: 4GB minimum, 8GB+ recommended
• Note: Older GPUs (RX 5000 series) may have slower performance

Intel GPUs (Vulkan):

• Supported: Intel Arc series (A750, A770, etc.)
• Requires: Latest Intel Graphics drivers with Vulkan support (Download)
• Performance: Moderate (~2-3s for 5s speech with Base model)
• VRAM: 8GB+ recommended

No GPU / Unsupported GPU (CPU Fallback):

• Performance: Slow (~5-10s for 5s speech with Base model)
• Recommended: Use Tiny or Base model for acceptable speed
• Works: Fully functional, just slower

VocalFold automatically detects and uses the best available runtime:

1. CUDA (NVIDIA GPUs) - Best performance
2. Vulkan (AMD/Intel GPUs) - Good performance
3. CPU (Fallback) - Slowest but functional

• .NET 9.0 SDK (for building from source)
• 16GB RAM (recommended)
• 8GB RAM (minimum)

1. Download VocalFold.exe from the Releases page
2. Install GPU drivers (see GPU Support section above)
3. Run VocalFold.exe
4. First run will download the AI model (~150MB, one-time)
5. Application runs in system tray

# Clone repository
git clone https://github.com/yourusername/VocalFold.git
cd VocalFold

# Build project
dotnet restore
dotnet build

# Run application
dotnet run --project VocalFold

# Or build standalone executable
dotnet publish -c Release -r win-x64 --self-contained true -p:PublishSingleFile=true

1. Launch VocalFold (runs in system tray)
2. Click in any text field (Notepad, browser, Word, etc.)
3. Press Ctrl+Windows (or your configured hotkey)
4. Speak clearly into your microphone
5. Your words appear as text at the cursor position

1. Right-click the VocalFold tray icon
2. Click Settings
3. Configure in the web UI:
   • Change global hotkey
   • Adjust typing speed
   • Manage keyword replacements
   • Organize keywords into categories
   • Enable/disable voice input

Create shortcuts for frequently used text:

• Say "comma" → types ","
• Say "period" → types "."
• Say "email signature" → types your full email signature
• Say "code snippet" → types your code template

Organize keywords into categories for better management.

Symptoms: Slow transcription (5-10 seconds), console shows "CPU mode"

Solutions:

For NVIDIA Users:

1. Install NVIDIA CUDA Toolkit 12.x: https://developer.nvidia.com/cuda-downloads
2. Verify installation: Open command prompt, run nvcc --version
3. Restart VocalFold

For AMD Users:

1. Install latest AMD Adrenalin drivers: https://www.amd.com/en/support
2. Verify Vulkan support:
   • Download Vulkan SDK: https://vulkan.lunarg.com/
   • Run vulkaninfo to check availability
3. Ensure GPU is RX 6000 series or newer
4. Restart VocalFold

For Intel Users:

1. Install latest Intel Graphics drivers: https://www.intel.com/content/www/us/en/download-center/home.html
2. Verify Vulkan support (Arc series required)
3. Restart VocalFold

Use these benchmarks to verify GPU is working:

Test method:

1. Record 5 seconds of clear speech
2. Use Base model
3. Measure transcription time (shown in console)

Expected performance:

• NVIDIA RTX 3080: <1s
• NVIDIA RTX 3060: <1.5s
• AMD RX 6800 XT: <1.5s
• AMD RX 6700 XT: <2s
• AMD RX 5700 XT: <3s
• Intel Arc A750: <2.5s
• CPU (i7-10700K): 5-8s

If your performance is significantly worse, GPU acceleration may not be working correctly.

• Hotkey not working: Check for conflicts with other applications
• Microphone not detected: Check Windows sound settings
• Application crashes: Check logs, report issue on GitHub

• Language: F# (.NET 9.0)
• AI Engine: Whisper.NET with CUDA and Vulkan runtimes
• Audio: NAudio
• Input Simulation: InputSimulatorCore
• Web UI: F# Fable + React + TailwindCSS
• Web Server: Giraffe + ASP.NET Core

• ✅ All processing happens locally on your machine
• ✅ No audio data is sent to external services
• ✅ No recordings stored to disk
• ✅ No telemetry or tracking
• ✅ Offline operation (after initial model download)

Contributions are welcome! Please feel free to submit a Pull Request.

[Your License Here]

• Whisper.NET - .NET bindings for OpenAI's Whisper
• NAudio - Audio library for .NET
• OpenAI - Original Whisper model

Made with ❤️ using F#