Spleeter

Dev Environment

Supported OS:

• Ubuntu 20.04
• macOS Big Sur v11

To setup developer environment, the project requires following packages.

apt-get update && apt-get upgrade -y && apt-get autoremove -y

# Installation of general dependencies
apt-get install -y build-essential clang-format clang-tidy clangd git git-lfs wget curl gnupg openjdk-11-jdk openjdk-11-jre lcov

# Installation of FFMPEG
apt-get install -y libavcodec-dev libavformat-dev libavfilter-dev libavdevice-dev libswresample-dev libswscale-dev ffmpeg

Build System

This project uses bazel build system. To install, run following command or find documentation for installation on office site here.

For Linux/macOS systems,

# Installation
curl -fsSL https://bazel.build/bazel-release.pub.gpg | gpg --dearmor > bazel.gpg
mv bazel.gpg /etc/apt/trusted.gpg.d/
echo "deb [arch=amd64] https://storage.googleapis.com/bazel-apt stable jdk1.8" | tee /etc/apt/sources.list.d/bazel.list
apt-get update && apt-get install -y bazel

# Command Completion
echo "source /etc/bash_completion.d/bazel" >> ~/.bashrc

# Autoformat
wget https://github.com/bazelbuild/buildtools/releases/download/3.5.0/buildifier
chmod +x buildifier
mv buildifier /usr/bin

Docker

One can use docker container to use as dev environment,

Install docker tool from official site here

docker pull registry.gitlab.com/jinay1991/spleeter

TensorFlow

To build libtensorflow_cc.so

~/$ git clone https://github.com/tensorflow/tensorflow.git
~/$ git checkout v2.3.0
~/$ cd tensorflow
~tensorflow/$ bazel build -c opt --config=monolithic //tensorflow:libtensorflow_cc.so //tensorflow:install_headers
~tensorflow/$ cd bazel-bin/tensorflow
~tensorflow/bazel-bin/tensorflow$ mkdir -p libtensorflow_cc-2.3.0-linux/include libtensorflow_cc-2.3.0-linux/lib
~tensorflow/bazel-bin/tensorflow$ cp -R include/* libtensorflow_cc-2.3.0-linux/include
~tensorflow/bazel-bin/tensorflow$ cp -P libtensorflow_cc.so libtensorflow_cc.so.2 libtensorflow_cc.so.2.3.0 libtensorflow_cc-2.3.0-linux/lib/
~tensorflow/bazel-bin/tensorflow$ tar cvzf libtensorflow_cc-2.3.0-linux.tar.gz libtensorflow_cc-2.3.0-linux

Package libtensorflow_cc-2.3.0-linux.tar.gz contains required tensorflow libraries.

TensorFlow Lite

To build libtensorflowlite.so

~/$ git clone https://github.com/tensorflow/tensorflow.git
~/$ git checkout v2.3.0
~/$ cd tensorflow
~tensorflow/$ bazel build -c opt --config=monolithic //tensorflow/lite:libtensorflowlite.so
~tensorflow/$ cd bazel-bin/tensorflow/lite
~tensorflow/bazel-bin/tensorflow$ mkdir -p libtensorflowlite-2.3.0-linux/include libtensorflowlite-2.3.0-linux/lib
~tensorflow/bazel-bin/tensorflow$ cp --parents -r ~tensorflow/lite/**/**/*.h libtensorflowlite-2.3.0-linux/include
~tensorflow/bazel-bin/tensorflow$ cp -P libtensorflowlite.so libtensorflowlite-2.3.0-linux/lib/
~tensorflow/bazel-bin/tensorflow$ tar cvzf libtensorflowlite-2.3.0-linux.tar.gz libtensorflowlite-2.3.0-linux

Package libtensorflowlite-2.3.0-linux.tar.gz contains required tensorflowlite libraries.

Usage

To quickly run the spleeter application, run bazel run //application:spleeter

To quickly run unit/component tests, run bazel test //... --test_output=all --cache_test_results=false

To build package (*.tar.gz), run bazel build //:spleeter-dev (This contains all the required header and libraries from this repository for smooth integration to other projects. See #integration guide to understand it better.)

Integration

Use the tarball for integration which contains all the required headers and libraries along with the third_party dependencies which can be fetched using bazel. Please note that spleeter has some external dependencies (i.e. libtensorflow etc.), which is why third_party directory is necessary for downloading all the dependencies and link them together with your application.

Usage on API can be found on Doxygen documentation.

Example implementation can be found in example/spleeter_app.cpp (Snipate below)

///
/// @file
/// @copyright Copyright (c) 2020, MIT License
///
#include "spleeter/argument_parser/cli_options.h"
#include "spleeter/spleeter.h"

#include <iostream>
#include <memory>

int main(void)
{
    try
    {
        /// Initialize
        auto cli_options = spleeter::CLIOptions{};
        cli_options.inputs = std::string{"external/audio_example/file/audio_example.wav"};
        cli_options.output_path = std::string{"separated_audio"};
        cli_options.configuration = std::string{"spleeter:5stems"};

        cli_options.audio_adapter = std::string{"audionamix"};
        cli_options.codec = std::string{"wav"};
        cli_options.bitrate = 192000;

        auto spleeter = std::make_unique<spleeter::Spleeter>(cli_options);
        spleeter->Init();

        /// Run
        spleeter->Execute();

        /// Deinitialize
        spleeter->Shutdown();
    }
    catch (std::exception& e)
    {
        std::cerr << "Caught Exception!! " << e.what() << std::endl;
        return 1;
    }

    return 0;
}

Prebuilt binaries for ubuntu-20.04-amd64 https://github.com/jinay1991/spleeter/releases/download/v1.4/spleeter-dev.tar.gz

Reference

• Deezer Research - Source Separation Engine Story - deezer.io blog post:
  • English version
  • Japanese version
• Music Source Separation tool with pre-trained models / ISMIR2019 extended abstract

If you use Spleeter in your work, please cite:

@misc{spleeter2019,
  title={Spleeter: A Fast And State-of-the Art Music Source Separation Tool With Pre-trained Models},
  author={Romain Hennequin and Anis Khlif and Felix Voituret and Manuel Moussallam},
  howpublished={Late-Breaking/Demo ISMIR 2019},
  month={November},
  note={Deezer Research},
  year={2019}
}

Converted official checkpoint to TFLite Model using https://github.com/tinoucas/spleeter-tflite-convert

License

The code of Spleeter is MIT-licensed.

Disclaimer

If you plan to use Spleeter on copyrighted material, make sure you get proper authorization from right owners beforehand.

Note

This repository include a demo audio file audio_example.mp3 which is an excerpt from

Slow Motion Dream by Steven M Bryant
Copyright (c) 2011 Licensed under a Creative Commons Attribution (3.0) license.

http://dig.ccmixter.org/files/stevieb357/34740
Ft: CSoul Alex Beroza & Robert Siek"