Musly

(c) 2013-2014, Dominik Schnitzer [email protected] and 2014-2016, Jan Schlüter [email protected]

Musly is a program and library for high performance audio music similarity computation. Musly only uses the audio signal when computing similarities! No meta data is used in the similarity computation process.

The source code is released under the MPL 2.0 see the file <musly/COPYING>

• The main Musly website is located at: http://www.musly.org
• Please fork me at: https://github.com/dominikschnitzer/musly
• Please report any Musly bug at: https://github.com/dominikschnitzer/musly/issues
• Documentation is found in the local <musly/doc> directory and on the main website.

Version History

VERSION 0.2 (under development)

Not released yet.

The second public Musly release. It includes the following changes and additions:

• musly_jukebox_addtracks() requires a fifth argument generate_ids: If nonzero, sequential track IDs are generated by Musly (as before), and if zero, track IDs are given by the user.
• musly_jukebox_removetracks(), musly_jukebox_trackcount(), musly_jukebox_maxtrackid(), and `musly_jukebox_gettrackids() are added to the API.
• musly_jukebox_guessneighbors_filtered() is added to the API, allowing the filter step to be constrained to a subset of registered tracks.
• musly_track_analyze_audiofile() accepts two parameters excerpt_length and excerpt_start controlling the length and position of the audio excerpt to be decoded and analyzed instead of the previous max_seconds, and the libav-based decoder's performance has been improved significantly.
• The musly command line client can compute sparse distance matrices (-s).
• When configured with -DUSE_OPENMP=On in cmake (or enabling the option in ccmake), the musly command line client and library are built with OpenMP support. The command line client parallelizes several computations.
• musly_jukebox_binsize(), musly_jukebox_tobin(), musly_jukebox_frombin(), musly_jukebox_tostream(), musly_jukebox_fromstream(), musly_jukebox_tofile() and musly_jukebox_fromfile() are added to the API, allowing to serialize and restore the internal state of a jukebox influenced by calls to musly_jukebox_setmusicstyle(), musly_jukebox_addtracks() and musly_jukebox_removetracks(). The command line client makes use of this when given the option -j or -J.
• musly_findmin() is added to the API.

VERSION 0.1

Released on 30 Jan 2014.

The first public Musly release. It includes two basic audio music similarity measures:

• mandelellis (as published in "M. Mandel and D. Ellis: Song-level features and support vector machines for music classification. In the proceedings of the 6th International Conference on Music Information Retrieval, ISMIR, 2005)

• timbre the mandelellis similarity measure tweaked for best results. We use a 25 MFCCs representation, the Jensen-Shannon divergence and normalize the similarities with Mutual Proximity (D. Schnitzer et al.: Using mutual proximity to improve content-based audio similarity. In the proceedings of the 12th International Society for Music Information Retrieval Conference, ISMIR, 2011).

For benchmarks and qualitative comparisons between other audio music similarity measures visit http://www.musly.org.

Installation

Musly uses the CMake build system, and depends on Eigen 3 and ffmpeg or libav 0.8 or above.

Ubuntu prerequisites

On Ubuntu, all requirements can be installed with: sudo apt-get install cmake libeigen3-dev libavcodec-dev libavformat-dev libavutil-dev

Windows prerequisites

For Windows, the easiest way seems to be to use MSYS2. Install it and launch the "MinGW-w64 Win64 Shell" from the start menu (not the "MSYS2 Shell"), then run:

# Update repository information
pacman -Sy
# Install prerequisites
pacman -S mingw-w64-x86_64-{gcc,cmake-git,pkgconf,eigen3} make wget p7zip
# Download and install precompiled ffmpeg libraries
wget https://ffmpeg.zeranoe.com/builds/win64/dev/ffmpeg-latest-win64-dev.7z
wget https://ffmpeg.zeranoe.com/builds/win64/shared/ffmpeg-latest-win64-shared.7z
7z x ffmpeg-latest-win64-dev.7z
7z x ffmpeg-latest-win64-shared.7z
mv ffmpeg-latest-win64-dev/include/* /usr/include
mv ffmpeg-latest-win64-dev/lib/* /usr/lib
mv ffmpeg-latest-win64-shared/bin/* /usr/bin
rm -rf ffmpeg-latest-win64-*
# Download and extract latest musly snapshot from git
wget https://github.com/dominikschnitzer/musly/archive/master.zip
7z x master.zip
cd musly-master

Compilation

To compile Musly, it is good practice to create an empty directory to build in. This can be placed inside your Musly source tree. Assuming you are in the source tree, run the following commands to build and install Musly (Note: For Windows, add -G"MSYS Makefiles" to both of the cmake lines below):

mkdir -p build && cd build  # create and switch into build directory
cmake ..                    # configure Musly
make                        # compile Musly
sudo make install           # install Musly

To perform a self-test of the library, optionally run (still inside build):

cmake -DBUILD_TEST=ON ..
make
# cp -a libmusly/*.dll libmusly/*/*.dll test/  # required for Windows only
ctest -V

It should end with 100% tests passed, 0 tests failed.

Command Line Tool

The command line interface is able to:

• Generate M3U playlists for music collections.
• Output full or sparse similarity matrices for more in-depth research of the audio music similarity functions. It uses the music-ir.org MIREX format (see <musly/doc/MIREX-DistanceMatrix.md>)
• Additionally the music similarity features can be ouput in text format to ease reuse of the features.

The command line tool is called "musly". Use "musly -h" to read about all available options. See http://www.musly.org for more information.

Library

Please see the library documentation in <doc/html> directory or on-line.

Development

To help Musly development and fix bugs the most convenient way is to use Eclipse. To generate an Eclipse project run:

cmake -G"Eclipse CDT4 - Unix Makefiles"\
      -DCMAKE_ECLIPSE_EXECUTABLE=/path/to/eclipse/eclipse \
      /path/to/musly-src

from an empty directory. The directory is filled with Eclipse project information which can be imported as an Eclipse project.