Voice to prompt, best for vibe coding

English | 中文

Watch the demo to see Talkie-Codie in action:

A cross-platform desktop application that converts your voice into optimized prompts using AI-powered speech recognition and language model enhancement.

• Python 3.10 or higher
• Microphone access

Before running the application, you may need to install system-level audio libraries.

If you encounter "PortAudio library not found" errors, please follow the audio setup guide:

📖 Audio Dependencies Setup Guide

1. Optional: Create virtual environment

   # Using conda
   conda create -n talkie-codie python=3.10
   conda activate talkie-codie

2. CUDA Support (Optional) If you have an NVIDIA GPU and want to use CUDA acceleration, manually install pytorch accordingly (for example, my GPU is sm 120 architecture):

   pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128

3. Launch GUI (auto-installs dependencies)

   python run_gui.py

Note: First startup may be slow due to model downloads. Configure your API key in the GUI settings.

Note: Without API configuration, the app will only use Whisper for audio-to-text conversion.

All settings can be configured through the GUI:

• LLM Provider & API Key: OpenAI or DeepSeek
• Whisper Settings: Device, model size, compute type
• Audio Device: Select your microphone

python src/main.py

1. Select Input Device: Choose your microphone from the dropdown
2. Start Recording: Click the record button to begin voice capture
3. View Results: See your transcribed text and AI-enhanced prompt
4. Copy Output: Use the copy button to copy the optimized prompt
5. Settings: Access configuration options via the settings button

# View cache information
python scripts/clear_cache.py info

# Clear cache (with confirmation)
python scripts/clear_cache.py clear

# Force clear cache
python scripts/clear_cache.py clear-force

• PyQt6: Modern GUI framework
• sounddevice: Audio recording and playback
• faster-whisper: Speech-to-text transcription
• scipy/numpy: Scientific computing
• requests: HTTP client for API calls

1. No audio input detected

   • Check microphone permissions
   • Verify device selection in settings
   • Ensure microphone is not muted

2. LLM API errors

   • Verify API key is correct
   • Check internet connection
   • Ensure sufficient API credits

3. Whisper model download issues

   • Check internet connection
   • Verify sufficient disk space
   • Try different model size in settings

Note: This application requires an active internet connection for LLM API calls and initial Whisper model downloads.