demucs-web

A web interface for Demucs, a state-of-the-art music source separation model. Separate vocals, drums, bass and others from any track.

Deployment

First, you need to deploy the Demucs server inside demucs folder somewhere. This server will be responsible for separating the audio tracks, so it needs GPU support.

I deployed this on Lightning AI.

These are environment variables for the Demucs server:

• API_URL: URL of the web interface server
• MODEL: Model to use for separation. htdemucs_ft is recommended. Read more about the models here.

Then you can deploy the web interface server:

docker compose up -d --build

These are environment variables for the web interface server:

• SECRET: Secret key for the server, used for authentication
• DEMUCS_API: URL of the Demucs server
• DEMUCS_API_KEY: API key for the Demucs server (optional, if deployed on Lightning AI)

Development

Make sure you have Bun installed, then run:

bun dev

Check http://localhost:3000 to see the result.

License

This project is licensed under the MIT license. See the LICENSE file for more information.