TuneSplit

An AI driven, GPU-accelerated stem separation webapp based on demucs

Usage

Note

You'll need docker to run this application

Using docker (Recommended)

As of version 2.1.0, TuneSplit is published as a ready-to-go, batteries included image to docker hub.

To quickly get started you can use the following docker-compose.yml:

services:
  app:
    image: rdelorenzo/tunesplit:latest
    environment:
      - ENVIRONMENT=production  # just silences logs
    ports:
      - "3003:3003"
    volumes:
      - models:/root/.cache/torch/hub/checkpoints
    restart: always

volumes:
  models:

For more details on the image, including more detailed instructions and even how to enable GPU support, check the image overview on docker hub.

Build from source

Use npm run docker to build and run the docker env (with docker-compose)

This will expose the app on localhost:3003

Run in deployment (CI USE)

npm run docker-deploy

Development

Run the full app for development (reloading)

npm run start-dev

Run the frontend for development (HMR)

npm run dev

Build

npm run build

Run the app as in production

npm run start-prod

Committing

git cz

Logging

Use log(message) instead of console.log(message)

Roadmap

• [x] Auto cleanup (maybe when download is done?)
• [x] Add auto-restart of container (and auto-start on boot)
• [x] Error handling to backend and frontend
• [x] Set up docker volumes
• [x] Set up CI/CD

NGINX Configuration

You must add the following properties in the locationscope:

client_max_body_size 0;

proxy_connect_timeout 300;
proxy_send_timeout 300;
proxy_read_timeout 300;
send_timeout 300;

Technologies

• node (@18.x)
• express
• react
• bash
• docker & docker-compose
• GitHub CI/CD
• Spleeter (by Deezer Research)