README.md

metadata

title: Music Separator Space
emoji: 💻
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 5.34.1
app_file: app.py
pinned: false
license: mit

🎵 Music Source Separation Demo

This Gradio Space lets you upload any mono .wav music mix and instantly separate it into drums, bass, other, and vocals stems using a pretrained U-Net model.

👉 Try it live: https://huggingface.co/spaces/theadityamittal/music-separator-space

---

🛠️ How It Works

1. Upload your mixture .wav (16 kHz mono).
2. Internally, we compute an STFT and split it into fixed-length segments.
3. A U-Net predicts magnitude masks for each source.
4. We reconstruct each stem via inverse STFT using the original phase.
5. Download your separated stems!

---

🔧 Installation (Local)

If you want to run it locally instead of on the Space:

git clone https://huggingface.co/spaces/theadityamittal/music-separator-space
cd music-separator-space
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
python serve.py

Then open http://localhost:7860 in your browser.

---

🗂️ Repository Structure

.
├── serve.py               # Gradio app entrypoint
├── requirements.txt       # Python dependencies
├── config/
│   └── default.yaml       # sample_rate, n_fft, hop_length, segment_length
└── src/
    └── models/
        └── unet.py        # U-Net architecture definition

---

⚙️ Configuration

• Model checkpoint is automatically downloaded from the Model Hub:

  https://huggingface.co/theadityamittal/music-separator-unet
  

• Environment variable CHECKPOINT_PATH may override the default path.

---

📄 License

This demo and the model checkpoint are released under the MIT license.

---

📝 Citation

If you use this demo or the model in your work, please cite:

@misc{music-separator-unet2025,
  title   = {Music Source Separation with U-Net},
  author  = {{Your Name}},
  year    = {2025},
  howpublished = {\url{https://huggingface.co/<YOUR_USERNAME>/music-separator-unet}},
}