Note that packages location for Python installed via Homebrew is different from the system Python. If you plan to use Essentia with Python, make sure the Homebrew directory is at the top of your PATH environment variable. To this end, add the line:
at the bottom of your
~/.bash_profile file. More information about using Python and Homebrew is here.
We are currently preparing deb packages for Ubuntu and Debian. Meanwhile, you need to compile Essentia from source (see below).
Windows, Android, iOS¶
Cross-compile Essentia from Linux/OSX (see below).
Compiling Essentia from source¶
- Essentia depends on (at least) the following libraries:
- FFTW: for the FFT implementation (optional)
- libavcodec/libavformat/libavutil/libavresample (from the FFmpeg/LibAv project): for loading/saving any type of audio files (optional)
- libsamplerate: for resampling audio (optional)
- TagLib: for reading audio metadata tags (optional)
- LibYAML: for YAML files input/output (optional)
- Gaia: for using SVM classifier models (optional)
- Chromaprint: for audio fingerprinting (optional)
All dependencies are optional, and some functionality will be excluded when a dependency is not found.
Installing dependencies on Linux¶
You can install those dependencies on a Debian/Ubuntu system from official repositories using the command below:
sudo apt-get install build-essential libyaml-dev libfftw3-dev libavcodec-dev libavformat-dev libavutil-dev libavresample-dev python-dev libsamplerate0-dev libtag1-dev libchromaprint-dev python-six
In order to use python bindings for the library, you might also need to install python-dev, python-numpy-dev (or python-numpy on Ubuntu) and python-yaml for YAML support in python:
sudo apt-get install python-dev python-numpy-dev python-numpy python-yaml
Similarly, in the case of Python3 install:
sudo apt-get install python3-dev python3-numpy-dev python3-numpy python3-yaml
Note that, depending on the version of Essentia, different versions of libav* and libtag1-dev packages are required. See release notes for official releases.
Since the 2.1-beta3 release of Essentia, the required version of TagLib (libtag1-dev) is greater or equal to
1.9. The required version of LibAv (
libavresample-dev) is greater or equal to
10. The appropriate versions are distributed in Ubuntu 14.10 or later, and in Debian wheezy-backports. If you want to install Essentia on older versions of Ubuntu/Debian, you will have to install a proper LibAv version from source.
Installing dependencies on Mac OS X¶
Install Command Line Tools for Xcode. Even if you install Xcode from the app store you must configure command-line compilation by running:
Install Homebrew package manager.
Insert the Homebrew directory at the top of your PATH environment variable by adding the following line at the bottom of your
brew install pkg-config gcc readline sqlite gdbm freetype libpng
Install Essentia’s dependencies:
brew install libyaml fftw ffmpeg libsamplerate libtag
Install python environment using Homebrew (Note that you are advised to do as described here and there are good reasons to do so. You will most probably encounter installation errors when using python/numpy preinstalled with OSX.):
brew install python --framework pip install ipython numpy matplotlib pyyaml
- Once your dependencies are installed, you can proceed to compiling Essentia. Download Essentia’s source code at Github. Due to different dependencies requirement (see release notes for official releases), make sure to download the version compatible with your system:
- 2.1 beta3 is the version currently recommended to install. It is supported on Ubuntu 14.10 or later, Debian Jessie or later and OSX. Build LibAv from source for support on Ubuntu 14.04 LTS or Debian Wheezy.
- master branch is the most updated version of Essentia in development
Go into its source code directory and start by configuring it:
./waf configure --build-static --with-python --with-cpptests --with-examples --with-vamp
- Use these (optional) flags:
--with-pythonto enable python bindings,
--with-examplesto build command line extractors based on the library,
--with-vampto build Vamp plugin wrapper,
--with-gaiato build with Gaia library support
--mode=debugto build in debug mode
--with-cppteststo build cpptests
NOTE: you must always configure at least once before building!
The following will give you the full list of options:
To compile everything you’ve configured:
All built examples will be located in
build/src/examples/ folder, as well as the Vamp plugin file
To install the C++ library, python bindings, extractors and Vamp plugin (if configured successfully; you might need to run this command with sudo):
Compiling for Python3¶
The waf build scripts are python scripts themselves. They will configure Essentia to be used with the same Python that was used to execute them. In the case if your default python is not Python3, you will need to run all waf commands with python3:
python3 ./waf configure --build-static --with-python --with-cpptests --with-examples --with-vamp python3 ./waf python3 ./waf install
Running tests (optional)¶
If you want to assure that Essentia works correctly, do the tests. Some of the tests require additional audio files, which are stored in a separate submodule repository essentia-audio. Make sure to clone Essentia git repository including its submodules in order to be able to run the tests (
git clone --recursive https://github.com/MTG/essentia.git).
To run the C++ base unit tests (only test basic library behavior):
To run the python unit tests (include all unittests on algorithms, need python bindings installed first):
or, in the case if your default python is not Python3:
python3 ./waf run_python_tests
Building documentation (optional)¶
All documentation is provided on the official website of Essentia library. Follow the steps below to generate it by yourself.
Install doxigen and pip, if you are on Linux:
sudo apt-get install doxygen python-pip
Install additional dependencies (you might need to run this command with sudo):
sudo pip install sphinx pyparsing sphinxcontrib-doxylink docutils jupyter sphinxprettysearchresults sudo apt-get install pandoc
Make sure to install Essentia with python bindings and run:
Documentation will be located in
Building Essentia on Windows¶
Essentia C++ library and extractors based on it can be compiled and run correctly on Windows, but python bindings are not supported yet. The easiest way to build Essentia is by cross-compilation on Linux using MinGW. However the resulting library binaries are only compatible within C++ projects using MinGW compilers, and therefore they are not compatible with Visual Studio. If you want to use Visual Studio, there is no project readily available, so you will have to setup one yourself and compile the dependencies too.
Building Essentia on Windows 10 via Bash on Ubuntu¶
It is possible to install Essentia easily via bash on Ubuntu on Windows 10. Bash on Ubuntu allows to run the same command-line utilities that could be run within a native Ubuntu 14.04 environment. Note that Bash on Ubuntu is still a beta product, hence there are some missing features and several issues. In addition, you cannot call Windows applications from bash.
To install bash on Ubuntu, follow the official guide in the Microsoft Developer Network.
After bash on Ubuntu is successfully installed, you should open a bash terminal and install the dependencies (see: Installing dependencies on Linux). Remember that bash on Windows runs on an Ubuntu 14.04 environment. Therefore, you may need to install a proper LibAv version from source.
Finally, you can compile Essentia (see: Compiling Essentia from source).
Building Essentia on Android¶
A lightweight version of Essentia can be cross-compiled for Android from Linux or Mac OSX.
Building Essentia on iOS¶
A lightweight version of Essentia can be cross-compiled for iOS from Mac OSX.
Using pre-trained high-level models in Essentia¶
Essentia includes a number of pre-trained classifier models for genres, moods and instrumentation. In order to use them you need to:
- Install Gaia2 library (supported on Linux/OSX)
- Build Essentia with examples and Gaia (
essentia_streaming_extractor_music(see detailed documentation)
You can train your own classifier models.