Table of contents:

• Purpose
• Updates
• Description
• Download data
• How to cite
• Contribution of new data
• Acknowledgements
• Links to other resources

Purpose

This database was created to help to evaluate of Musical Audio Signal Separation algorithms and statements on a representative set of professionally produced music (i.e. real music recordings).

Updates

(31/12/2008)

• Tracks of Sargon-Silenci were rearranged.
• All audio files were normalized to 44.1Khz and 24bits.

Description

• Tracks (with/without effects): Stereo Microsoft PCM WAV files (44.1Khz, 24 bits) of every instrumental track including and/or without including effects (plugins enabled or disabled in the project file used for production)
• Description of the effects: When available, we include a description of the plugins used to modify the tracks without effects.
• Lyrics: When available, lyrics are included.
• License: All licenses allow these data sets to be used in scientific research but other may grant other priviledges (such as by-nc Creative Common licensed work).

Instructions for loading a data set in matlab

1. Uncompress the data set in a folder
2. Download the following matlab file: loadSources.zip
3. Execute in matlab as follows:
   

   [sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
   = loadSources(datasetFolder,effects);

   With effects set to 0 to load the tracks without effects and set to 1 to load the tracks with effects.

Download data

List of artist-songs (genre):

• sargon-silenci (metal)
• bearlin-roads (pop-rock)
• tamy-que_pena_tanto_faz (bossanova)
• fort_minor-remember_the_name (hip-hop)
• vieux_farka_toure-ana (rock-reggae-blues)
• kismet-tv_on (indie-rock)

[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('sargon-silenci_1-19',1);
[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('sargon-silenci_1-19',0);

[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('sargon-silenci_22-37',1);
[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('sargon-silenci_22-37',0);

[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('bearlin-roads_14-42',1);
[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('bearlin-roads_14-42',0);

[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('bearlin-roads_99-113',1);
[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('bearlin-roads_99-113',0);

[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('tamy-que_pena_tanto_faz_6-19',1);

[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('tamy-que_pena_tanto_faz_46-57',1);

[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('fort_minor-remember_the_name_127-145',1);

[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('fort_minor-remember_the_name_199-204',1);

[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('vieux_farka_toure-ana_30-55',0);

[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('vieux_farka_toure-ana_201-213',0);

[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('kismet-tv_on_0-24',0);

[sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] 
= loadSources('kismet-tv_on_100-128',0);

How to cite

@MISC{MTGMASSdb,
      author = {Vinyes, MarC},
      title = {{MTG MASS database}},
      howpublished = {\\\url{http://www.mtg.upf.edu/static/mass/resources}},
      year = {2008},
}

Contribution of new data

This website doesn't accept contributions at this time, however we encourage other people to create similar data sets. For this purpose we would like to share some points that, in our experience, are important when you ask a professional producer for these materials:

• All audio files of one song have to be time aligned. Consequently, tracks with and without effects should be rendered with the same location markers in the project file used for production.
• All sounds should be rendered using the most frequent same sample rate and number of bits per sample used for recording, avoiding as much resampling or requantization steps as posssible.
• An "effect", as a modification of an acoustic sound, is an ill-defined concept when you're used to cut, add, duplicate, loop and record sounds from all kinds of electronic devices. It also may happen that the sound is already recorded with effects when hardware effects are used: for example electric guitars are usually recorded from an amp. Here we decided to call "tracks without effects" those tracks rendered with all the extra digital effects plugins disabled except from the panning and volume automations, and we call "tracks with effects" those tracks that are rendered with everything enabled. This is the way we told producers to render these materials.
• Finally, it is important to make sure that tracks with and without effects are rendered as stereo files so that stereo effects like panning, stereo delays and reverberations are effectively represented in the data.

At the end we also scaled uniformly (with a constant gain for each song) all of its tracks so that their mixture had maximum amplitude without clipping in a Microsoft PCM WAV file.

Acknowledgements

These materials were prepared by MarC Vinyes Raso. We would like to thank Bearlin and Sargon for releasing the tracks of their songs and Sergi Vila at Garatge Produccions and Juan Pedro Barroso at Kcleta Studios for their effort to provide us what we requested.

This work was developed in the Music Technology Grop and supported by the SAME project.

Use our contact page if you wish to contact us.

Links to other resources

• ENST-Drums: an extensive audio-visual database for drum signals processing.
• Nine Inch Nails remix website (wav).
• ccMixter: a community music site featuring remixes licensed under Creative Commons (wav/mp3).
• Acid planet remixing contests.
• INRIA BASS-dB: the Blind Audio Source Separation evaluation database (mp3).