Professionally produced music recordings


See the results webpage (external link)

Test Data

Tamy - Que pena tanto faz (snip2) (external link)
License (external link)

Bearlin - Roads (snip2) (external link)
License (external link)

The data consist of stereo WAV audio files, that can be imported in Matlab using the wavread command:
  • [mix,fs,nbits]=wavread('tamy-que_pena_tanto_faz_sisec08.wav'); mixL=mix(:,1); mixR=mix(:,2);
  • [mix,fs,nbits]=wavread('bearlin-roads_sisec08.wav'); mixL=mix(:,1); mixR=mix(:,2);

Development Data

Mixture and tracks from different snips of the same songs.

Tamy - Que pena tanto faz (snip1) (external link)
License (external link)

Bearlin - Roads (snip1) (external link)
License (external link)

Instructions to load the tracks (sources) and the mixture in matlab:
  • Download matlab code: (external link)
  • Uncompress all zip files to the same folder
  • Execute:
    • [sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] = loadSources('tamy-que_pena_tanto_faz_6-19',1);
    • [sourcesL,sourcesR,sourceNameList,mixL,mixR,fs,nbits] = loadSources('bearlin-roads_85-99',1);

You can use other songs from the Musical Audio Signal Separation Evaluation Resources (external link) of the Music Technology Group (MTG) at Universitat Pompeu Fabra.


Tamy - Que pena tanto faz (snip2)
Extract the following stereo tracks:
  • vocals
  • guitar

Bearlin - Roads (snip2)
Extract the following stereo tracks:
  • bass
  • vocals
  • piano


Participants may submit separation results for one or both of the above mixtures.

In addition, each participant is asked to provide basic information about his/her algorithm (e.g. a bibliographical reference) and to declare its average running time, expressed in seconds per test excerpt and per GHz of CPU.

Note that the submitted audio files will be made available on a website under the terms of the Creative Commons Attribution-NonCommercial-ShareAlike 2.0 (external link) license.

Evaluation criteria

The same basic evaluation criteria as for the under-determined speech and music mixtures dataset will be used first so that results are comparable. More precisely, the estimated stereo source signals will be evaluated via the criteria used for the Stereo Audio Source Separation Evaluation Campaign (external link), except that the order of the sources is fixed. These criteria distinguish spatial (or filtering) distortion, interference and artifacts.

Additionally the Signal To Error Ratio from the Magnitude Spectrograms of the estimated source and the error will be computed over the left and right stereo channels. The spectrograms are built using a Blackman Harris -92dB window, frames of 4096 samples chosen every 1024 samples (hop size).
  • Associated matlab code: (external link)
  • Execute: specMagnitudeSER_L=SISECerrorsSpectrogram(wavSource(:,1),wavEstSource(:,1),selectWindow(4096,5,3),1024) specMagnitudeSER_R=SISECerrorsSpectrogram(wavSource(:,2),wavEstSource(:,2),selectWindow(4096,5,3),1024)

Performance will be compared to that of ideal binary masking as a benchmark (i.e. binary masks providing maximum SDR), computed over a STFT or a cochleagram.

Potential participants

M. Vinyes
Vasileios Pantazis

Task proposed by: M. Vinyes