Musly - an opensource audio music similarity library

Musly is a fast and high-quality audio music similarity library written in C/C++. It currently implements two music similarity algorithms. A third commercial one can be licensed from OFAI. The implemented similarity routines are described and evaluated in more depth in the Similarity Methods page.

Musly is licensed under the terms of the MPL 2.0 open source license, a permissive weak copyleft license.

Musly can be used as library in your program or through a small command line application. The command line application

General information about Audio Music Similarity can be found in my PhD thesis which is also available as book on Amazon.

Requirements

Installation

Download the latest release from Github.
Extract it to a local directory (SRCDIR)
Create an empty directory to compile Musly (BINDIR)
Execute cmake SRCDIR while in BINDIR to configure Musly
Execute make && make install to compile and install Musly
Read the Documentation to understand how you can use the Musly library in your program.

Usage of the Command Line Client

Initialize a new Musly collection

# musly -N

Analyze all MP3 files in the given path

# musly -x mp3 -a /path/to/music

Compute a playlist

# musly -p metal.00047.au
Computing the k=5 most similar tracks to: metal.00047.au
  metal.00053.au
  metal.00054.au
  metal.00056.au
  metal.00050.au
  metal.00049.au

All options

Music Similarity Library (Musly) - http://www.musly.org
Version: 0.1
(c) 2013-2014, Dominik Schnitzer 

Options for musly/musly/musly
  -h           this help screen.
  -v 0-5       set the libmusly debug level: (0: none, 5: trace).
               DEFAULT: 0
  -i           information about the music similarity library
  -c COLL      set the file to write the music similarity features and
               to use for computing similarities.
               DEFAULT: collection.musly
  -k NUM       set number of similar songs display when computing
               playlists ('-p') or evaluating the collection ('-e').
               DEFAULT: 5
 INITIALIZATION:
  -n MTH | -N  initialize the collection (set with '-c') using the
               music similarity method MTH. Available methods:
               mandelellis,timbre
               '-N' automatically selects the best method.
 MUSIC ANALYSIS/PLAYLIST GENERATION:
  -a DIR/FILE  analyze and add the given audio FILE to the collection
               file. If a Directory is given, the directory is scanned
               recursively for audio files.
  -x EXT       only analyze files with file extension EXT when adding
               audio files with '-a'. DEFAULT: '' (any)
  -p FILE      print a playlist of the '-k' most similar tracks for
               the given FILE. If FILE is not found in the collection
               file, it is analyzed and then compared to all other
               tracks found in the collection file ('-c').
 LISTING:
  -l           list all files in the collection file.
  -d           dump the features in the collection file to the console
 EVALUATION:
  -e NUM | -E  perform a basic kNN (k-nearest neighbor) music genre
               classification experiment using the selected collection
               file. The parameter k is set with option '-k'. The
               genre is inferred from the path element at position NUM.
               The genre position within the path is guessed with '-E'.
  -f NUM       Use an artist filter for the evaluation ('-e'). The 
               artist name is inferred from the path element at
               position NUM.
               DEFAULT: -1 (No artist filter)
  -m FILE      compute the full similarity matrix for the specified
               collection and write it to FILE. It is written in MIREX
               text format (see http://www.music-ir.org/mirex).

Authors

Dominik Schnitzer

About