SAM-Audio Isolation Utility

A local audio isolation utility powered by Meta's SAM-Audio model, optimized for Apple Silicon via MLX.

Features

• 🎵 Text-prompted audio isolation - Describe what you want to extract (speech, vocals, drums, piano, etc.)
• 🖥️ Modern web interface - Dark theme with drag-and-drop upload
• 🍎 Apple Silicon optimized - Uses MLX for fast inference on M Series Macs
• 📦 Easy setup - Single script handles everything

Screenshot

Requirements

• Hardware: Apple Silicon Mac (M Series)
• RAM: 16GB minimum (64GB recommended for best performance)
• macOS: 14.0 or later
• Python: 3.10 or later (auto-downloaded if missing)

Python Setup

This application requires Python 3.10 or later. If your Mac only has the system Python, run.sh will automatically download a private Python runtime into .python/ and use it (no system changes).

Auto-download requirements:

• curl available (built into macOS)
• Internet access
• ~150MB free disk space

To force a re-download, delete the .python/ directory and run ./run.sh again.

Quick Start

# Clone the repository
git clone https://github.com/cmd-christopher/Mac-GUI-for-SAM-Audio.git
cd Mac-GUI-for-SAM-Audio

# Run the application (creates venv, installs deps, downloads model)
./run.sh

Then open the URL printed in the terminal (starts at http://localhost:5001 and auto-adjusts if that port is busy).

Note: On first run, the model (~4.8GB) will be downloaded automatically. This is a one-time process.

Usage

1. Upload an audio file (MP3, WAV, FLAC, M4A, OGG, AAC)
2. Describe the sound to isolate (e.g., "speech", "piano", "drums")
3. Click "Isolate Audio"
4. Listen to the isolated audio and residual
5. Download the results as WAV files

Example Prompts

Prompt	Isolates
speech	Human voices
vocals	Singing in music
drums	Percussion
piano	Piano instrument
guitar	Guitar sounds
music	All musical elements

Project Structure

Mac-GUI-for-SAM-Audio/
├── app.py                 # Flask web application
├── audio_processor.py     # SAM-Audio model wrapper
├── requirements.txt       # Python dependencies
├── run.sh                 # Startup script
├── templates/
│   └── index.html         # Web interface
└── static/
    ├── css/style.css      # Styling
    └── js/app.js          # Frontend logic

Technical Details

• Model: mlx-community/sam-audio-large-fp16
• Framework: MLX (Apple's machine learning framework)
• Backend: Flask
• Sample Rate: 24kHz
• Output Format: WAV

Troubleshooting

"Python 3.10+ is required" error

The automatic Python download failed or was blocked. Re-run ./run.sh with a working internet connection, or delete .python/ and try again.

Port 5001 in use

The app automatically selects the next available port and prints it at startup. To force a specific port, run:

SAM_AUDIO_PORT=5005 ./run.sh

Model download fails

Ensure you have a stable internet connection. The model (~4.8GB) is downloaded from Hugging Face.

Out of memory

Enable "Long audio mode" checkbox for files over 30 seconds - this uses chunked processing.

"resolution-too-deep" pip error

This usually means the bundled environment didn't finish installing dependencies. Delete .python/ and venv/, then run ./run.sh again.

Credits

• SAM-Audio by Meta AI
• MLX by Apple
• mlx-audio by Lucas Newman

License

MIT License - See LICENSE for details.