Skip to content

JeffreyCA/spleeter-web

Repository files navigation

Spleeter Web

Latest release Commits since latest release Docker Compose push (master)

Spleeter Web is a web application for isolating or removing the vocal, accompaniment, bass, and/or drum components of any song. For example, you can use it to isolate the vocals of a track, or you can use it remove the vocals to get an instrumental version of a song.

It supports a number of different source separation models: Spleeter (4stems-model and 5stems-model), Demucs, CrossNet-Open-Unmix, and D3Net.

The app uses Django for the backend API and React for the frontend. Celery is used for the task queue. Docker images are available, including ones with GPU support.

Table of Contents

Features

  • Supports Spleeter, Demucs, CrossNet-Open-Unmix (X-UMX), and D3Net source separation models
    • Each model supports a different set of user-configurable parameters in the UI
  • Dynamic Mixes let you export and play back in realtime your own custom mix of the different components
  • Import tracks by uploading an audio file or by a YouTube link
    • Built-in YouTube search functionality (YouTube Data API key required)
  • Supports lossy (MP3) and lossless (FLAC, WAV) output formats
  • Persistent audio library with ability to stream and download your source tracks and mixes
  • Customize number of background workers working on audio separation and YouTube imports
  • Supports third-party storage backends like S3 and Azure Blob Storage
  • Clean and responsive UI
  • Support for GPU separation
  • Fully Dockerized

Homepage

Upload modal

Mixer

Getting started with Docker

Requirements

Instructions

  1. Clone repo:

    $ git clone https://github.com/JeffreyCA/spleeter-web.git
    $ cd spleeter-web
  2. (Optional) Set the YouTube Data API key (for YouTube search functionality):

    You can skip this step, but you would not be able to import songs by searching with a query. You would still be able to import songs via YouTube links though.

    Create an .env file at the project root with the following contents:

    YOUTUBE_API_KEY=<YouTube Data API key>
    
  3. (Optional) Setup for GPU support: Source separation can be accelerated with a GPU (however only NVIDIA GPUs are supported).

    1. Install NVIDIA drivers for your GPU.

    2. Install the NVIDIA Container Toolkit. If on Windows, refer to this.

    3. Verify Docker works with your GPU by running sudo docker run --rm --gpus all nvidia/cuda:11.6.2-base-ubuntu20.04 nvidia-smi

  4. Download and run prebuilt Docker images:

    Note: On Apple Silicon and other AArch64 systems, the Docker images need to be built from source.

    # CPU separation
    spleeter-web$ docker-compose -f docker-compose.yml -f docker-compose.prod.yml -f docker-compose.prod.selfhost.yml up
    # GPU separation
    spleeter-web$ docker-compose -f docker-compose.gpu.yml -f docker-compose.prod.yml -f docker-compose.prod.selfhost.yml up

    Alternatively, you can build the Docker images from source:

    # CPU separation
    spleeter-web$ docker-compose -f docker-compose.yml -f docker-compose.build.yml -f docker-compose.prod.yml -f docker-compose.prod.selfhost.yml up --build
    # GPU separation
    spleeter-web$ docker-compose -f docker-compose.gpu.yml -f docker-compose.build.gpu.yml -f docker-compose.prod.yml -f docker-compose.prod.selfhost.yml up --build
  5. Launch Spleeter Web

    Navigate to http://127.0.0.1:80 in your browser. Uploaded tracks and generated mixes will appear in media/uploads and media/separate respectively on your host machine.

Getting started without Docker

If you are on Windows, it's recommended to follow the Docker instructions above. Celery is not well-supported on Windows.

Requirements

  • x86-64 arch (For AArch64 systems, use Docker)
  • 4 GB+ of memory (source separation is memory-intensive)
  • Python 3.8+ (link)
  • Node.js 16+ (link)
  • Redis (link)
  • ffmpeg and ffprobe (link)
    • On macOS, you can install it using Homebrew or MacPorts
    • On Windows, you can follow this guide

Instructions

  1. Set environment variables

    Make sure these variables are set in every terminal session prior to running the commands below.

    # Unix/macOS:
    (env) spleeter-web$ export YOUTUBE_API_KEY=<api key>
    # Windows:
    (env) spleeter-web$ set YOUTUBE_API_KEY=<api key>
  2. Create Python virtual environment

    spleeter-web$ python -m venv env
    # Unix/macOS:
    spleeter-web$ source env/bin/activate
    # Windows:
    spleeter-web$ .\env\Scripts\activate
  3. Install Python dependencies

    (env) spleeter-web$ pip install -r requirements.txt
    (env) spleeter-web$ pip install -r requirements-spleeter.txt --no-dependencies
  4. Install Node dependencies

    spleeter-web$ cd frontend
    spleeter-web/frontend$ npm install
  5. Ensure Redis server is running on localhost:6379 (needed for Celery)

    You can run it on a different host or port, but make sure to update CELERY_BROKER_URL and CELERY_RESULT_BACKEND in settings.py. It must be follow the format: redis://host:port/db.

  6. Apply migrations

    (env) spleeter-web$ python manage.py migrate
  7. Build frontend

    spleeter-web$ npm run build --prefix frontend
  8. Start backend in separate terminal

    (env) spleeter-web$ python manage.py collectstatic && python manage.py runserver 127.0.0.1:8000
  9. Start Celery workers in separate terminal

    Unix/macOS:

    # Start fast worker
    (env) spleeter-web$ celery -A api worker -l INFO -Q fast_queue -c 3
    
    # Start slow worker
    (env) spleeter-web$ celery -A api worker -l INFO -Q slow_queue -c 1

    This launches two Celery workers: one processes fast tasks like YouTube imports and the other processes slow tasks like source separation. The one working on fast tasks can work on 3 tasks concurrently, while the one working on slow tasks only handles a single task at a time (since it's memory-intensive). Feel free to adjust these values to your fitting.

    Windows:

    You'll first need to install gevent. Note however that you will not be able to abort in-progress tasks if using Celery on Windows.

    (env) spleeter-web$ pip install gevent
    # Start fast worker
    (env) spleeter-web$ celery -A api worker -l INFO -Q fast_queue -c 3 --pool=gevent
    
    # Start slow worker
    (env) spleeter-web$ celery -A api worker -l INFO -Q slow_queue -c 1 --pool=gevent
  10. Launch Spleeter Web

    Navigate to http://127.0.0.1:8000 in your browser. Uploaded and mixed tracks will appear in media/uploads and media/separate respectively.

Configuration

Django settings

Settings file Description
django_react/settings.py The base Django settings used when launched in non-Docker context.
django_react/settings_dev.py Contains the override settings used when run in development mode (i.e. DJANGO_DEVELOPMENT is set).
django_react/settings_docker.py The base Django settings used when launched using Docker.
django_react/settings_docker_dev.py Contains the override settings used when run in development mode using Docker (i.e. docker-compose.dev.yml).

Environment variables

Here is a list of all the environment variables you can use to further customize Spleeter Web:

Name Description
CPU_SEPARATION No need to set this if using Docker. Otherwise, set to 1 if you want CPU separation and 0 if you want GPU separation.
DJANGO_DEVELOPMENT Set to true if you want to run development build, which uses settings_dev.py/settings_docker_dev.py and runs Webpack in dev mode.
ALLOW_ALL_HOSTS Set to 1 if you want Django to allow all hosts, overriding any APP_HOST value. This effectively sets the Django setting ALLOWED_HOSTS to [*]. There are security risks associated with doing this. Default: 0
APP_HOST Domain name(s) or public IP(s) of server. To specify multiple hosts, separate them by a comma (,).
API_HOST Hostname of API server (for nginx).
DEFAULT_FILE_STORAGE Whether to use local filesystem or cloud-based storage for storing uploads and separated files. FILE or AWS or AZURE.
AWS_ACCESS_KEY_ID AWS access key. Used when DEFAULT_FILE_STORAGE is set to AWS.
AWS_SECRET_ACCESS_KEY AWS secret access key. Used when DEFAULT_FILE_STORAGE is set to AWS.
AWS_STORAGE_BUCKET_NAME AWS S3 storage bucket name. Used when DEFAULT_FILE_STORAGE is set to AWS.
AWS_S3_CUSTOM_DOMAIN Custom domain, such as for a CDN. Used when DEFAULT_FILE_STORAGE is set to AWS.
AWS_S3_REGION_NAME S3 region (e.g. us-east-1). Used when DEFAULT_FILE_STORAGE is set to AWS.
AWS_S3_SIGNATURE_VERSION Default signature version used for generating presigned urls. To be able to access your s3 objects in all regions through presigned urls, set this to s3v4. Used when DEFAULT_FILE_STORAGE is set to AWS.
AZURE_ACCOUNT_KEY Azure Blob account key. Used when DEFAULT_FILE_STORAGE is set to AZURE.
AZURE_ACCOUNT_NAME Azure Blob account name. Used when DEFAULT_FILE_STORAGE is set to AZURE.
AZURE_CONTAINER Azure Blob container name. Used when DEFAULT_FILE_STORAGE is set to AZURE.
AZURE_CUSTOM_DOMAIN Custom domain, such as for a CDN. Used when DEFAULT_FILE_STORAGE is set to AZURE.
CELERY_BROKER_URL Broker URL for Celery (e.g. redis://localhost:6379/0).
CELERY_RESULT_BACKEND Result backend for Celery (e.g. redis://localhost:6379/0).
CELERY_FAST_QUEUE_CONCURRENCY Number of concurrent YouTube import tasks Celery can process. Docker only.
CELERY_SLOW_QUEUE_CONCURRENCY Number of concurrent source separation tasks Celery can process. Docker only.
CERTBOT_DOMAIN Domain for creating HTTPS certs using Let's Encrypt's Certbot. Docker only.
CERTBOT_EMAIL Email address for creating HTTPS certs using Let's Encrypt's Certbot. Docker only.
D3NET_OPENVINO Set to 1 to use OpenVINO for D3Net CPU separation. Requires Intel CPU.
DEMUCS_SEGMENT_SIZE Length of each split for GPU separation. Default is 40, which requires a around 7 GB of GPU memory. For GPUs with 2-4 GB of memory, experiment with lower values (minimum is 10). Also recommended to set PYTORCH_NO_CUDA_MEMORY_CACHING=1.
D3NET_OPENVINO_THREADS Set to the number of CPU threads for D3Net OpenVINO separation. Default: # of CPUs on the machine. Requires Intel CPU.
DEV_WEBSERVER_PORT Port that development webserver is mapped to on host machine. Docker only.
ENABLE_CROSS_ORIGIN_HEADERS Set to 1 to set Cross-Origin-Embedder-Policy and Cross-Origin-Opener-Policy headers which are required for exporting Dynamic Mixes in-browser.
NGINX_PORT Port that Nginx is mapped to on host machine for HTTP. Docker only.
NGINX_PORT_SSL Port that Nginx is mapped to on host machine for HTTPS. Docker only.
PYTORCH_NO_CUDA_MEMORY_CACHING Set to 1 to disable Pytorch caching for GPU separation. May help with Demucs separation on lower memory GPUs. Also see DEMUCS_SEGMENT_SIZE.
UPLOAD_FILE_SIZE_LIMIT Maximum allowed upload file size (in megabytes). Default is 100.
YOUTUBE_API_KEY YouTube Data API key.
YOUTUBE_LENGTH_LIMIT Maximum allowed YouTube track length (in minutes). Default is 30.
YOUTUBEDL_SOURCE_ADDR Client-side IP address for yt-dlp to bind to. If you are facing 403 Forbidden errors, try setting this to 0.0.0.0 to force all connections through IPv4.
YOUTUBEDL_VERBOSE Set to 1 to enable verbose logging for yt-dlp.

Using cloud storage (Azure Storage, AWS S3, etc.)

By default, Spleeter Web uses the local filesystem to store uploaded files and mixes. It uses django-storages, so you can also configure it to use other storage backends like Azure Storage or AWS S3.

You can set the environment variable DEFAULT_FILE_STORAGE (.env if using Docker) to either FILE (for local storage), AWS (S3 storage), or AZURE (Azure Storage).

Then, depending on which backend you're using, set these additional variables:

AWS S3:

  • AWS_ACCESS_KEY_ID
  • AWS_SECRET_ACCESS_KEY
  • AWS_STORAGE_BUCKET_NAME

Azure Storage:

  • AZURE_ACCOUNT_KEY
  • AZURE_ACCOUNT_NAME
  • AZURE_CONTAINER

CORS

To play back a dynamic mix, you may need to configure your storage service's CORS settings to allow the Access-Control-Allow-Origin header.

If you have ENABLE_CROSS_ORIGIN_HEADERS set, then you'll need to additionally set the Cross-Origin-Resource-Policy response headers of audio files to cross-origin. See this for more details.

Deployment

Spleeter Web can be deployed on a VPS or a cloud server such as Azure VMs, AWS EC2, DigitalOcean, etc. Deploying to cloud container services like ECS is not yet supported out of the box.

  1. Clone this git repo

    $ git clone https://github.com/JeffreyCA/spleeter-web.git
    $ cd spleeter-web
  2. (Optional) If self-hosting, update docker-compose.prod.selfhost.yml and replace ./media with the path where media files should be stored on the server.

  3. In spleeter-web, create an .env file with the production environment variables

    Example .env file:

    APP_HOST=<domain name(s) or public IP(s) of server>
    DEFAULT_FILE_STORAGE=<FILE or AWS or AZURE>       # Optional (default = FILE)
    CELERY_FAST_QUEUE_CONCURRENCY=<concurrency count> # Optional (default = 3)
    CELERY_SLOW_QUEUE_CONCURRENCY=<concurrency count> # Optional (default = 1)
    YOUTUBE_API_KEY=<youtube api key>                 # Optional
    

    See Environment Variables for all the available variables. You can also set these directly in the docker-compose.*.yml files.

  4. Build and start production containers

    For GPU separation, replace docker-compose.yml and docker-compose.build.yml below for docker-compose.gpu.yml and docker-compose.build.gpu.yml respectively.

    If you are self-hosting media files:

    # Use prebuilt images
    spleeter-web$ sudo docker-compose -f docker-compose.yml -f docker-compose.prod.yml -f docker-compose.prod.selfhost.yml up -d
    # Or build from source
    spleeter-web$ sudo docker-compose -f docker-compose.yml -f docker-compose.build.yml -f docker-compose.prod.yml -f docker-compose.prod.selfhost.yml up --build -d

    Otherwise if using a storage provider:

    # Use prebuilt images
    spleeter-web$ sudo docker-compose -f docker-compose.yml -f docker-compose.prod.yml up -d
    # Or build from source
    spleeter-web$ sudo docker-compose -f docker-compose.yml -f docker-compose.build.yml -f docker-compose.prod.yml up --build -d
  5. Access Spleeter Web at whatever you set APP_HOST to. Note that it will be running on port 80, not 8000. You can change this by setting NGINX_PORT and NGINX_PORT_SSL.

HTTPS support

Enabling HTTPS allows you to export Dynamic Mixes from your browser. To enable HTTPS, set both CERTBOT_DOMAIN and CERTBOT_EMAIL to your domain name and CERTBOT_EMAIL to your email in .env and include -f docker-compose.https.yml in your docker-compose up command.

Credits

Special thanks to my Sponsors:

And especially to all the researchers and devs behind all the source separation models:

And additional thanks to these wonderful projects:

Turntable icon made from Icon Fonts is licensed by CC BY 3.0.

License

MIT