Skip to main content
Glama

Stem MCP Server

by tolutronics
README.md32 kB
# Stem MCP Server 🎵 A comprehensive Model Context Protocol (MCP) server for professional AI-powered audio processing and stem manipulation. Designed specifically for music producers, audio engineers, and Logic Pro users who need advanced audio processing capabilities integrated with AI workflows. ## Features 🚀 ### 🎯 Core Audio Processing - **🎤 AI Stem Generation**: State-of-the-art source separation using Demucs models - **✂️ Smart Audio Splitting**: Intelligent segmentation with customizable overlap and fade options - **🔄 Seamless Loop Creation**: Professional loop generation with tempo matching and crossfading - **📊 Advanced Audio Analysis**: Deep musical feature extraction (tempo, key, spectral characteristics) - **🎯 Precise Instrument Isolation**: Extract specific instruments with multiple algorithms - **🎵 Vocal Processing**: Advanced vocal extraction and separation techniques ### 🎛️ Advanced Features - **🎪 Multi-Vocal Range Separation**: Split vocals into soprano, alto, tenor, bass ranges - **🎼 Musical Structure Analysis**: Detect beats, tempo, key signatures, and harmonic content - **🔊 Dynamic Range Analysis**: RMS energy, peak detection, loudness analysis - **🎚️ Spectral Processing**: Frequency domain analysis and manipulation - **⚡ Batch Processing**: Handle multiple files efficiently - **🎨 Custom Processing Chains**: Combine multiple tools for complex workflows ### Supported Audio Formats - WAV, MP3, FLAC, AAC, M4A, OGG, WMA ### AI Models - **Demucs**: State-of-the-art source separation models - `htdemucs` (default): High-quality 4-stem separation - `htdemucs_ft`: Fine-tuned variant - `htdemucs_6s`: 6-stem separation - `mdx`: Alternative model architecture - `mdx_extra`: Enhanced MDX model ## Installation 🔧 ### Prerequisites - Python 3.10 or higher (required for MCP compatibility) - FFmpeg (for audio processing) - CUDA-compatible GPU (optional, for faster processing) ### Install FFmpeg ```bash # macOS (using Homebrew) brew install ffmpeg # Ubuntu/Debian sudo apt update && sudo apt install ffmpeg # Windows (using Chocolatey) choco install ffmpeg ``` ### Install the MCP Server ```bash # Clone or create the project cd stem-mcp # Install in development mode pip install -e . # Or install from requirements pip install -r requirements.txt ``` ### Install Dependencies ```bash # Install core dependencies pip install mcp>=1.0.0 librosa soundfile numpy scipy torch torchaudio demucs pydub # For best performance, install with CUDA support pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu118 ``` ## Configuration ⚙️ ### MCP Client Configuration Add this to your MCP client configuration (e.g., Claude Desktop): ```json { "mcpServers": { "stem-processing": { "command": "stem-mcp", "args": [], "env": { "PYTHONPATH": "/path/to/stem-mcp/src" } } } } ``` ## Usage Examples 🎯 ### 1. Generate Stems from Audio ```bash # Using the MCP tool { "tool": "generate_stems", "arguments": { "audio_path": "/path/to/song.wav", "output_dir": "./stems", "model_type": "htdemucs", "num_stems": 4 } } ``` **Output**: Separates audio into vocals, drums, bass, and other instruments. ### 2. Split Stems into Segments ```bash { "tool": "split_stems", "arguments": { "stem_path": "./stems/vocals.wav", "output_dir": "./segments", "segment_length": 15.0, "overlap": 2.0 } } ``` **Output**: Creates 15-second segments with 2-second overlap. ### 3. Create Seamless Loops ```bash { "tool": "create_loop", "arguments": { "audio_path": "./drums.wav", "loop_duration": 8.0, "bpm": 120, "crossfade_duration": 0.2 } } ``` **Output**: Creates an 8-second loop at 120 BPM with smooth crossfading. ### 4. Analyze Audio Features ```bash { "tool": "analyze_audio", "arguments": { "audio_path": "./song.wav" } } ``` **Output**: ``` 🎵 Audio Analysis Results: 📊 Basic Info: Duration: 245.67 seconds Sample Rate: 44100 Hz Channels: Stereo 🎵 Musical Features: Tempo: 128.5 BPM Estimated Key: G Beat Count: 523 🔊 Spectral Analysis: Avg Spectral Centroid: 2847.3 Hz Avg Spectral Rolloff: 8934.2 Hz Avg Zero Crossing Rate: 0.0847 Avg RMS Energy: 0.1234 ``` ### 5. Extract Vocals Only ```bash { "tool": "extract_vocal", "arguments": { "audio_path": "./song.wav", "method": "demucs" } } ``` ### 6. Isolate Specific Instruments ```bash { "tool": "isolate_instrument", "arguments": { "audio_path": "./song.wav", "instrument": "drums", "method": "demucs" } } ``` ## API Reference 📚 ### Complete Tool Suite #### 🎤 `generate_stems` State-of-the-art AI-powered source separation using Demucs models. **Parameters:** - `audio_path` (string, required): Path to input audio file - `output_dir` (string, optional): Output directory (default: ".") - `model_type` (string, optional): Demucs model type - `"htdemucs"` (default): High-quality 4-stem separation - `"htdemucs_ft"`: Fine-tuned variant for enhanced quality - `"htdemucs_6s"`: 6-stem separation (vocals, drums, bass, piano, guitar, other) - `"mdx"`: Fast processing with good quality - `"mdx_extra"`: Enhanced MDX model - `num_stems` (integer, optional): Number of output stems (2-6, default: 4) **Output**: Generates separate audio files for each stem (vocals, drums, bass, other) --- #### ✂️ `split_stems` Intelligent audio segmentation with customizable parameters. **Parameters:** - `stem_path` (string, required): Path to audio file to split - `output_dir` (string, optional): Output directory (default: ".") - `segment_length` (number, optional): Segment duration in seconds (1-300, default: 30) - `overlap` (number, optional): Overlap between segments in seconds (0-10, default: 0) **Features:** - Smart segment boundary detection - Customizable overlap for smooth transitions - Preserves audio quality and metadata --- #### 🔄 `create_loop` Professional seamless loop creation with advanced crossfading. **Parameters:** - `audio_path` (string, required): Path to input audio - `output_path` (string, optional): Output file path (auto-generated if not provided) - `loop_duration` (number, optional): Loop duration in seconds (0.5-60, default: 4) - `bpm` (number, optional): Target BPM (60-200, auto-detected if not specified) - `crossfade_duration` (number, optional): Crossfade length in seconds (0-2, default: 0.1) **Features:** - Automatic tempo detection and matching - Smart beat-aligned loop points - Professional crossfading algorithms - Maintains musical timing and feel --- #### 📊 `analyze_audio` Comprehensive musical and spectral analysis. **Parameters:** - `audio_path` (string, required): Path to audio file to analyze **Analysis Output:** - **Basic Properties**: Duration, sample rate, channel configuration - **Musical Features**: Tempo (BPM), key signature, beat tracking - **Spectral Analysis**: Frequency content, spectral centroid, rolloff - **Dynamic Range**: RMS energy levels, peak detection - **Audio Quality**: Zero-crossing rate, harmonic content --- #### 🎤 `extract_vocal` Advanced vocal extraction with multiple algorithms. **Parameters:** - `audio_path` (string, required): Path to input audio - `output_path` (string, optional): Output file path (auto-generated if not provided) - `method` (string, optional): Extraction algorithm - `"demucs"` (default): AI-powered high-quality separation - `"librosa"`: Traditional signal processing approach - `"spectral"`: Frequency domain processing **Features:** - Multiple extraction algorithms for different use cases - High-quality vocal isolation - Preserves vocal character and dynamics --- #### 🎹 `isolate_instrument` Precise instrument isolation using multiple techniques. **Parameters:** - `audio_path` (string, required): Path to input audio - `instrument` (string, optional): Target instrument - `"vocals"`: Lead and backing vocals - `"drums"`: Full drum kit - `"bass"`: Bass guitar and synthesizers - `"guitar"`: Electric and acoustic guitars - `"piano"`: Piano and keyboard instruments - `"other"`: Remaining instruments - `output_path` (string, optional): Output file path - `method` (string, optional): Isolation technique - `"demucs"`: AI source separation - `"librosa"`: Signal processing - `"spectral"`: Frequency domain filtering --- #### 🎪 `separate_vocal_ranges` **NEW**: Advanced vocal range separation for choir and multi-vocal arrangements. **Parameters:** - `audio_path` (string, required): Path to vocal audio file - `output_dir` (string, optional): Output directory for range files **Output**: Separate files for each vocal range: - **Soprano**: High female voices (C4-C6) - **Alto**: Low female voices (G3-E5) - **Tenor**: High male voices (C3-A4) - **Bass**: Low male voices (E2-E4) **Features:** - Frequency-based intelligent separation - Preserves natural vocal characteristics - Ideal for choir arrangements and vocal analysis --- #### 🎵 `extract_vocal_harmonies` **NEW**: Isolate and separate vocal harmonies from lead vocals. **Parameters:** - `audio_path` (string, required): Path to audio with vocal harmonies - `output_dir` (string, optional): Directory for harmony files - `sensitivity` (number, optional): Harmony detection sensitivity (0.1-1.0, default: 0.5) **Features:** - Separates lead vocals from harmonies - Maintains harmonic relationships - Perfect for remixing and vocal arrangement analysis ## Performance Tips 🚀 ### Hardware Optimization - **GPU**: Use CUDA-compatible GPU for 10x faster processing - **RAM**: 16GB+ recommended for processing large files - **Storage**: SSD recommended for faster I/O operations ### Processing Tips - **File Format**: Use WAV or FLAC for best quality - **Sample Rate**: 44.1kHz or 48kHz for optimal results - **Batch Processing**: Process multiple files in sequence for efficiency ### Model Selection - **htdemucs**: Best general-purpose model - **htdemucs_6s**: Use for 6-stem separation (vocals, drums, bass, piano, guitar, residual) - **mdx**: Faster processing, slightly lower quality ## Development 😠️ ### 💻 Complete Project Structure ``` stem-mcp/ ├── src/stem_mcp/ │ ├── __init__.py # Package initialization and version │ ├── server.py # Main MCP server implementation │ ├── audio_processor.py # Core audio processing engine │ ├── tools_schema.py # MCP tool definitions and schemas │ ├── utils.py # Utility functions and helpers │ ├── vocal_processor.py # Advanced vocal processing tools │ └── analysis_engine.py # Audio analysis and feature extraction ├── examples/ │ ├── test_tools.py # Comprehensive tool testing script │ ├── sample_workflows.py # Example production workflows │ └── integration_examples.py # Logic Pro integration examples ├── tests/ │ ├── test_audio_processing.py # Audio processing tests │ ├── test_vocal_tools.py # Vocal processing tests │ ├── test_analysis.py # Analysis engine tests │ └── test_integration.py # MCP integration tests ├── docs/ │ ├── API_REFERENCE.md # Detailed API documentation │ ├── WORKFLOWS.md # Production workflow guides │ ├── TROUBLESHOOTING.md # Common issues and solutions │ └── PERFORMANCE_GUIDE.md # Optimization tips and benchmarks ├── pyproject.toml # Project configuration and dependencies ├── requirements.txt # Python dependencies ├── requirements-dev.txt # Development dependencies ├── .gitignore # Git ignore patterns ├── DEMO_COMPLETE.md # Complete demo and feature overview ├── README.md # This comprehensive guide └── LICENSE # MIT License ``` ### 🔧 Development Environment Setup #### **Quick Start** ```bash # Clone the repository git clone <repository-url> cd stem-mcp # Create virtual environment python -m venv venv source venv/bin/activate # On Windows: venv\Scripts\activate # Install in development mode with all dependencies pip install -e ".[dev]" # Install additional development tools pip install -r requirements-dev.txt # Verify installation stem-mcp --version ``` #### **Development Dependencies** ```bash # Core development tools pip install pytest pytest-cov black flake8 mypy pre-commit # Audio testing tools pip install pytest-audio librosa-test-utils # Performance profiling pip install memory-profiler line-profiler # Documentation tools pip install sphinx sphinx-rtd-theme ``` ### 🚀 Running in Development Mode #### **Basic Development Commands** ```bash # Run server with debug logging stem-mcp --debug --log-level DEBUG # Run with specific configuration stem-mcp --config config/dev_config.json # Run with performance profiling stem-mcp --profile --profile-output profile_results.txt # Test all tools with sample audio python examples/test_tools.py # Run comprehensive test suite pytest tests/ -v --cov=src/stem_mcp # Run specific test categories pytest tests/test_audio_processing.py -v pytest tests/test_vocal_tools.py -v ``` #### **Code Quality & Formatting** ```bash # Format code with Black black src/ tests/ examples/ # Lint with flake8 flake8 src/ tests/ examples/ # Type checking with mypy mypy src/stem_mcp/ # Run all quality checks pre-commit run --all-files ``` ### 🧪 Testing & Quality Assurance #### **Test Categories** - **Unit Tests**: Individual function and class testing - **Integration Tests**: MCP client-server communication - **Audio Tests**: Audio processing accuracy and quality - **Performance Tests**: Speed and memory usage benchmarks - **Regression Tests**: Ensure consistent outputs across versions #### **Running Tests** ```bash # Run all tests with coverage pytest tests/ --cov=src/stem_mcp --cov-report=html # Run tests with audio samples pytest tests/ --with-audio-samples # Run performance benchmarks pytest tests/test_performance.py --benchmark-only # Run memory usage tests pytest tests/test_memory.py --memray ``` ### 🔍 Debugging & Profiling #### **Debug Mode Features** - Detailed logging at all processing stages - Audio processing step visualization - Memory usage tracking - Processing time measurements - Model loading and caching information #### **Performance Profiling** ```bash # Profile CPU usage python -m cProfile -o profile.stats examples/test_tools.py # Profile memory usage python -m memory_profiler examples/test_tools.py # Profile specific functions @profile def my_function(): # Function code here pass ``` ### 🤝 Contributing Guidelines #### **Development Workflow** 1. **Fork the repository** and create your feature branch 2. **Set up development environment** with all dependencies 3. **Write comprehensive tests** for your changes 4. **Follow code style guidelines** (Black, flake8, mypy) 5. **Update documentation** for new features 6. **Run full test suite** before submitting 7. **Submit pull request** with detailed description #### **Code Style Standards** - **Python**: Follow PEP 8 with Black formatting - **Docstrings**: Google-style docstrings for all public functions - **Type Hints**: Use type hints for all function parameters and returns - **Comments**: Clear, concise comments for complex logic - **Error Handling**: Comprehensive error handling with informative messages #### **Pull Request Checklist** - ☑️ All tests pass locally - ☑️ Code follows style guidelines - ☑️ Documentation is updated - ☑️ New features have tests - ☑️ No breaking changes (or clearly documented) - ☑️ Performance impact assessed - ☑️ Example usage provided ## Professional Workflows 🎯 ### 🎚️ Logic Pro Integration Seamlessly integrate with Logic Pro for enhanced music production: #### **Complete Production Workflow** 1. **🎵 Export from Logic Pro** - Export stereo mix or individual tracks - Use 24-bit/48kHz for best quality - Export as WAV or AIFF format 2. **🤖 AI-Powered Processing** - Generate high-quality stems using Demucs - Analyze musical content and structure - Extract specific instruments or vocal parts - Create seamless loops from any section 3. **🎹 Import Back to Logic** - Import processed stems as individual tracks - Use analyzed BPM data for tempo matching - Apply extracted loops to new compositions - Layer isolated instruments for creative arrangements #### **Advanced Production Techniques** **🎭 Stem-Based Remixing** ```bash # 1. Generate stems from your Logic Pro export generate_stems("/path/to/logic_export.wav", model_type="htdemucs_6s") # 2. Analyze each stem for musical content analyze_audio("/stems/vocals.wav") analyze_audio("/stems/drums.wav") # 3. Create custom loops from specific sections create_loop("/stems/drums.wav", loop_duration=8, bpm=128) # 4. Extract vocal harmonies for detailed editing extract_vocal_harmonies("/stems/vocals.wav") ``` **🎵 Vocal Production Chain** ```bash # Complete vocal processing workflow extract_vocal("/audio/full_mix.wav", method="demucs") separate_vocal_ranges("/vocals/extracted_vocal.wav") extract_vocal_harmonies("/vocals/extracted_vocal.wav") ``` **🎶 Loop Library Creation** ```bash # Create a comprehensive loop library split_stems("/stems/drums.wav", segment_length=8, overlap=1) create_loop("/segments/drums_segment_001.wav", loop_duration=4) create_loop("/segments/bass_segment_002.wav", loop_duration=8) ``` --- ### 🎼 Music Production Use Cases #### **🎵 For Producers** - **Stem Analysis**: Understand song structure and arrangement - **Remixing**: Extract and manipulate individual elements - **Sample Creation**: Generate unique samples from existing tracks - **Loop Building**: Create custom loops for new productions #### **🎤 For Vocalists & Vocal Coaches** - **Vocal Isolation**: Extract clean vocal tracks from mixes - **Harmony Analysis**: Study vocal arrangements and harmonies - **Range Training**: Separate and analyze different vocal ranges - **Performance Analysis**: Study vocal techniques and patterns #### **🎸 For Musicians** - **Instrument Learning**: Isolate specific instruments for practice - **Transcription**: Extract clear instrument tracks for notation - **Performance Study**: Analyze playing techniques and arrangements - **Cover Creation**: Create backing tracks by removing specific instruments #### **🎧 For Audio Engineers** - **Mix Analysis**: Understand frequency content and arrangement - **Mastering Reference**: Compare individual stems and their processing - **Problem Solving**: Isolate problematic elements in complex mixes - **Quality Control**: Analyze audio content and detect issues --- ### 🔀 Complete Integration Example **Scenario**: Converting a Logic Pro song into stems for remixing ```bash # Step 1: Export your Logic Pro project as a stereo mix # File -> Export -> Audio... -> 24-bit WAV # Step 2: Generate high-quality stems generate_stems("/path/to/my_song.wav", model_type="htdemucs_6s", # 6-stem separation output_dir="./my_song_stems") # Step 3: Analyze each stem for musical information analyze_audio("./my_song_stems/vocals.wav") analyze_audio("./my_song_stems/drums.wav") analyze_audio("./my_song_stems/bass.wav") # Step 4: Create loops from interesting drum sections split_stems("./my_song_stems/drums.wav", segment_length=16, # 16-second segments overlap=2) # 2-second overlap # Step 5: Generate seamless loops create_loop("./segments/drums_segment_001.wav", loop_duration=8, crossfade_duration=0.5) # Step 6: Process vocals for detailed editing separate_vocal_ranges("./my_song_stems/vocals.wav") extract_vocal_harmonies("./my_song_stems/vocals.wav") # Step 7: Import all processed audio back to Logic Pro # - Drag stems into Logic as individual tracks # - Use loops in Logic's loop browser # - Apply vocal range files for detailed vocal editing ``` **Result**: Complete stem-based workflow with: - ✅ Individual instrument tracks - ✅ Seamless loops ready for new compositions - ✅ Separated vocal ranges for detailed editing - ✅ Extracted harmonies for remix work - ✅ Complete musical analysis data ## Advanced Troubleshooting 🔧 ### 🚫 Common Issues & Solutions #### **Installation Problems** **"ModuleNotFoundError: No module named 'demucs'"** ```bash # Install missing dependencies pip install demucs torch torchaudio # For CUDA support (recommended) pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu118 ``` **"FFmpeg not found"** ```bash # macOS brew install ffmpeg # Ubuntu/Debian sudo apt update && sudo apt install ffmpeg # Windows choco install ffmpeg # Or download from https://ffmpeg.org/download.html ``` **"MCP server not recognized"** ```bash # Ensure proper installation pip install -e . # Verify entry point stem-mcp --version # Check MCP client configuration cat ~/.config/claude-desktop/config.json ``` #### **Performance Issues** **"CUDA out of memory"** ```bash # Solution 1: Reduce memory usage # Process shorter segments split_stems("/large_file.wav", segment_length=30) # Solution 2: Use CPU processing # Set environment variable export CUDA_VISIBLE_DEVICES="" # Solution 3: Use lighter models generate_stems("/file.wav", model_type="mdx") # Faster, less memory # Solution 4: Clear GPU cache import torch torch.cuda.empty_cache() ``` **"Slow processing speeds"** ```bash # Install CUDA-optimized PyTorch pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu118 # Use optimal audio formats # Convert to WAV 44.1kHz before processing ffmpeg -i input.mp3 -ar 44100 -ac 2 output.wav # Use faster models for real-time needs generate_stems("/file.wav", model_type="mdx") # ~3x faster ``` **"High memory usage"** ```bash # Monitor memory usage import psutil print(f"Memory: {psutil.virtual_memory().percent}%") # Process in batches for segment in split_stems("/large_file.wav", segment_length=60): process_segment(segment) # Memory is freed after each segment ``` #### **Audio Quality Issues** **"Poor separation quality"** ```bash # Use highest quality models generate_stems("/file.wav", model_type="htdemucs_ft") # Ensure high-quality input # Use lossless formats (WAV, FLAC) when possible # Avoid heavily compressed MP3s # Pre-process audio for optimal results ffmpeg -i input.mp3 -ar 48000 -ab 320k output.wav ``` **"Artifacts in output"** ```bash # Check input file quality analyze_audio("/suspicious_file.wav") # Use different models for different content # Electronic music: "mdx_extra" # Live recordings: "htdemucs" # Vocals: "htdemucs_ft" ``` **"Loops don't sound seamless"** ```bash # Increase crossfade duration create_loop("/drums.wav", crossfade_duration=0.5) # Longer crossfade # Ensure tempo-aligned segments create_loop("/drums.wav", bpm=120, loop_duration=8) # 2-bar loop at 120 BPM # Analyze source material first analysis = analyze_audio("/drums.wav") print(f"Detected BPM: {analysis['tempo']}") ``` #### **File Format Issues** **"Unsupported audio format"** ```bash # Convert to supported format ffmpeg -i input.m4a -ar 44100 output.wav ffmpeg -i input.opus -ar 44100 output.wav # Batch convert multiple files for file in *.m4a; do ffmpeg -i "$file" "${file%.m4a}.wav" done ``` **"Audio file corrupted"** ```bash # Verify file integrity ffprobe -v error -show_entries stream=codec_name,duration -of csv=p=0 file.wav # Repair corrupted files ffmpeg -i corrupted.wav -c copy repaired.wav ``` ### 📝 Debugging Techniques #### **Enable Verbose Logging** ```python import logging logging.basicConfig(level=logging.DEBUG) # Or set environment variable log_level = "DEBUG" ``` #### **Audio Processing Diagnostics** ```python # Add debugging to your workflow result = analyze_audio("/test_file.wav") print(f"File info: {result}") # Check intermediate outputs stems = generate_stems("/test.wav", output_dir="./debug_stems") for stem in stems: analysis = analyze_audio(stem) print(f"{stem}: {analysis['duration']}s, {analysis['sample_rate']}Hz") ``` #### **Performance Monitoring** ```python import time import psutil def monitor_processing(): start_time = time.time() start_memory = psutil.virtual_memory().used # Your processing here result = generate_stems("/large_file.wav") end_time = time.time() end_memory = psutil.virtual_memory().used print(f"Processing time: {end_time - start_time:.2f}s") print(f"Memory used: {(end_memory - start_memory) / 1024**2:.2f}MB") ``` ### 🔍 Advanced Diagnostics #### **System Requirements Check** ```bash # Check Python version (3.10+ required) python --version # Check available memory free -h # Linux top -l 1 -n 0 | grep PhysMem # macOS # Check GPU availability python -c "import torch; print(f'CUDA available: {torch.cuda.is_available()}')" # Check disk space df -h ``` #### **Audio System Diagnostics** ```python import librosa import soundfile as sf # Test audio library functionality try: y, sr = librosa.load("/test.wav") print(f"LibROSA working: {len(y)} samples at {sr}Hz") except Exception as e: print(f"LibROSA error: {e}") try: data, samplerate = sf.read("/test.wav") print(f"SoundFile working: {len(data)} samples") except Exception as e: print(f"SoundFile error: {e}") ``` ## 📊 Performance Optimization Guide ### 🚀 Hardware Recommendations #### **Optimal System Configuration** - **CPU**: Intel i7/i9 or AMD Ryzen 7/9 (8+ cores recommended) - **RAM**: 32GB+ for professional use, 16GB minimum - **GPU**: NVIDIA RTX 3060+ with 8GB+ VRAM (for CUDA acceleration) - **Storage**: SSD for audio files (NVMe preferred for large files) - **OS**: Linux or macOS for best performance, Windows 11 supported #### **Performance Benchmarks** | Model Type | GPU (RTX 4090) | CPU (i9-12900K) | Memory Usage | |------------|----------------|-----------------|---------------| | htdemucs | ~45s (3min song) | ~180s | 6GB VRAM / 8GB RAM | | htdemucs_6s | ~60s (3min song) | ~240s | 8GB VRAM / 12GB RAM | | mdx | ~25s (3min song) | ~90s | 4GB VRAM / 6GB RAM | | mdx_extra | ~30s (3min song) | ~120s | 5GB VRAM / 8GB RAM | ### ⚡ Optimization Strategies #### **Model Selection Guide** ```python # For speed (real-time applications) generate_stems("/file.wav", model_type="mdx") # For quality (studio production) generate_stems("/file.wav", model_type="htdemucs_ft") # For versatility (6 stems including piano/guitar) generate_stems("/file.wav", model_type="htdemucs_6s") # For balanced speed/quality generate_stems("/file.wav", model_type="htdemucs") ``` #### **Batch Processing Optimization** ```python # Process multiple files efficiently audio_files = ["/song1.wav", "/song2.wav", "/song3.wav"] for audio_file in audio_files: # Reuse loaded model for better performance stems = generate_stems(audio_file, model_type="htdemucs") # Process each stem for stem in stems: analysis = analyze_audio(stem) # Store results for batch processing ``` #### **Memory Management** ```python import gc import torch def process_large_file(audio_path): # Clear GPU cache before processing if torch.cuda.is_available(): torch.cuda.empty_cache() # Process in segments for large files segments = split_stems(audio_path, segment_length=120) # 2-minute segments results = [] for segment in segments: result = generate_stems(segment) results.append(result) # Force garbage collection gc.collect() if torch.cuda.is_available(): torch.cuda.empty_cache() return results ``` --- ## 📚 Additional Resources ### 🎵 Music Production Resources - **Logic Pro User Guide**: Apple's official documentation - **Demucs Research Paper**: "Music Source Separation in the Waveform Domain" - **Audio Processing Theory**: Understanding digital signal processing - **MCP Specification**: Model Context Protocol documentation ### 🔗 Community & Support - **GitHub Issues**: Report bugs and request features - **Discussions**: Share workflows and get community help - **Discord**: Real-time chat with other users (coming soon) - **Blog**: Regular updates and tutorials (coming soon) ### 💰 Commercial Use This project is open source and free for both personal and commercial use under the MIT license. For enterprise support, custom integrations, or commercial licensing inquiries, please contact the maintainers. --- ## 📄 License MIT License Copyright (c) 2024 Stem MCP Server Contributors Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. --- ## 🙏 Acknowledgments ### 🎆 Core Technologies - **[Demucs](https://github.com/facebookresearch/demucs)**: State-of-the-art source separation by Meta Research - **[LibROSA](https://librosa.org/)**: Comprehensive audio analysis library - **[PyTorch](https://pytorch.org/)**: Deep learning framework powering AI models - **[MCP Protocol](https://modelcontextprotocol.io/)**: Model Context Protocol specification - **[SoundFile](https://github.com/bastibe/python-soundfile)**: Audio file I/O operations ### 🎵 Audio Processing Libraries - **FFmpeg**: Universal audio/video processing framework - **NumPy & SciPy**: Numerical computing foundations - **scikit-learn**: Machine learning utilities for audio analysis - **Pydub**: Simple audio manipulation toolkit ### 🔌 Integration Partners - **Logic Pro**: Apple's professional music production software - **Claude Desktop**: AI assistant with MCP support - **Music Production Community**: Producers, engineers, and musicians worldwide ### 👥 Contributors Thanks to all contributors who have helped make this project better: - Core development team - Beta testers and early adopters - Community feedback and feature requests - Documentation and example contributors ### 🏆 Special Recognition - **Meta Research**: For developing and open-sourcing Demucs - **Anthropic**: For creating the MCP protocol and supporting AI-audio workflows - **Apple**: For Logic Pro integration possibilities - **Open Source Community**: For the foundation libraries that make this possible --- ## 🎆 Project Stats - **📋 Languages**: Python (primary), Shell scripting - **📦 Dependencies**: 15+ core libraries, 50+ total with dev dependencies - **🤖 AI Models**: 5+ Demucs variants supported - **🎵 Audio Formats**: 8+ supported input/output formats - **⚙️ Tools**: 8+ MCP tools for comprehensive audio processing - **📊 Performance**: Up to 10x speed improvement with GPU acceleration - **🌍 Platform Support**: macOS, Linux, Windows --- <div align="center"> ## 🎵 **Happy Music Making!** 🎵 *Transform your audio with AI-powered precision* **[Get Started](#installation-🔧) | [View Examples](#usage-examples-🎯) | [Join Community](https://github.com/your-repo/discussions)** --- *Built with ♥️ for music producers, audio engineers, and creative professionals* **🎆 Powered by Demucs • 🤖 Enhanced by AI • 🎹 Designed for Logic Pro** </div>

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/tolutronics/audio-processing-mcp'

If you have feedback or need assistance with the MCP directory API, please join our Discord server