mkdir build
cd build
cmake .. -G "Visual Studio 16 2019" -A x64
cmake --build . --config Release

4. Generate Audio

./build/bin/stable_audio_cli \
    --model models/stable_audio.onnx \
    --prompt "120 BPM, 4 bar loop, electric piano, C major" \
    --output output.wav \
    --seconds 11 \
    --seed 42

📁 Project Structure

stable-audio-generator/
├── cpp/                    # C++ core library
│   ├── include/           # Public headers
│   ├── src/               # Implementation
│   ├── cli/               # Command-line tool
│   └── CMakeLists.txt     # Build configuration
├── scripts/               # Python utilities
│   └── convert_to_onnx.py # Model conversion script
├── models/                # ONNX models (generated, not in git)
├── third_party/           # ONNX Runtime (downloaded separately)
├── BUILD.md               # Detailed build instructions
├── README_CPP.md          # C++ version quick start
├── ARCHITECTURE.md        # Architecture documentation
└── CPP_MIGRATION_SUMMARY.md # Migration details

🔧 Configuration Options

Command Line

stable_audio_cli \
    --model models/stable_audio.onnx \
    --prompt "Your text prompt" \
    --output output.wav \
    --seconds 11 \
    --bars 4 \
    --bpm 120 \
    --key C \
    --seed 42 \
    --device cpu

Parameters:

• --model (required): Path to ONNX model file
• --prompt (required): Text description of desired audio
• --output: Output WAV file path (default: output.wav)
• --seconds: Length in seconds (default: 11.0)
• --bars: Number of musical bars (overrides seconds)
• --bpm: Beats per minute (default: 120)
• --key: Musical key (optional)
• --seed: Random seed (0 = random)
• --device: Device (cpu, cuda, coreml)

📱 Platform Support

🔍 Verification Steps

To verify Milestone 1 is working:

1. ✅ Build project: cmake --build .
2. ✅ Convert model: python scripts/convert_to_onnx.py
3. ✅ Generate test audio: ./stable_audio_cli --model models/stable_audio.onnx --prompt "test"
4. ✅ Verify output.wav is created and plays correctly

See verification.md for detailed acceptance criteria

📖 Documentation

• BUILD.md - Detailed build instructions for all platforms
• README_CPP.md - C++ version quick start guide
• ARCHITECTURE.md - Technical architecture details
• CPP_MIGRATION_SUMMARY.md - Migration from Python to C++
• cpp/README.md - C++ library API documentation

🐛 Troubleshooting

ONNX Runtime Not Found

• Ensure ONNX Runtime is extracted to third_party/onnxruntime-{platform}/
• Check that include/ and lib/ directories exist
• See BUILD.md for platform-specific setup

Model Conversion Fails

• Ensure Hugging Face authentication: huggingface-cli login
• Accept model license: https://huggingface.co/stabilityai/stable-audio-open-small
• Check Python dependencies are installed

Build Errors

• Verify CMake version: cmake --version (needs 3.18+)
• Check C++17 compiler is available
• See BUILD.md troubleshooting section

📝 License & Model Attribution

• This project code is provided under the license specified in the contract
• Stable Audio Open model is provided by Stability AI under their license
• Model must be downloaded and converted separately after accepting Hugging Face terms
• See LICENSE.md for full details

🆚 Why C++/ONNX?

This project uses C++ with ONNX Runtime instead of Python to achieve:

• ✅ True cross-platform: Same codebase for Windows, macOS, Linux, Android, iOS
• ✅ Native performance: No interpreter overhead
• ✅ Mobile support: Native Android/iOS builds
• ✅ Small binaries: No Python runtime needed
• ✅ Optimized inference: ONNX Runtime optimized for each platform

See CPP_MIGRATION_SUMMARY.md for details.

🤝 Support

For issues or questions:

1. Check BUILD.md for build issues
2. Review verification.md for common problems
3. Contact developer per contract terms

📦 Deliverables Checklist

• C++ project structure with CMake
• ONNX Runtime integration
• Command-line audio generation tool
• Model conversion script (Python)
• Cross-platform build configurations
• Documentation (build, architecture, API)
• Mobile proof-of-concept structure
• Model conversion completed (requires Hugging Face access)
• Mobile builds (Android/iOS)