Professionally produced music recordings


Introduction


The task includes the following data:

As compared to former SiSEC 2008 data (test1 and dev1), the new data (test2 and dev2) were prepared with following changes bringing the task closer to a "realistic situation":
  • Test and development data come from different recordings.
  • In addition to 20 seconds snips to be separated (test2 and dev2), full-length recordings (test2_full_mix and dev2_full_mix) are provided as well and can be used.

Results


Test Data


Download test1.zip (external link) (7 MB) (former test data of SiSEC 2008 professionally produced music recordings (external link) task)
Download test2.zip (external link) (9 MB) (newly proposed data)
Download test2_full_mix.zip (external link) (79 MB) (full-length recordings for newly proposed data)

The data consist of stereo WAV audio files, that can be imported in Matlab using the wavread command. These files are named {test1,test2}__[<author>]-[<song>]___[<snip>]__{mix,full_mix}.wav , where <author> is the author name, <song> is the song name, and <snip> is a shortcut for snip information.

The data include the following mixtures (snips and full-length recordings):

test1
  • test1__tamy-que_pena_tanto_faz__snip__mix.wav
  • test1__bearlin-roads__snip__mix.wav
test2
  • test2__glen_philips-the_spirit_of_shackleton__snip_163_185__mix.wav
  • test2__nine_inch_nails-the_good_soldier__snip_104_125__mix.wav
  • test2__shannon_hurley-sunrise__snip_62_85__mix.wav
test2_full_mix
  • test2__glen_philips-the_spirit_of_shackleton__full_mix.wav
  • test2__nine_inch_nails-the_good_soldier__full_mix.wav
  • test2__shannon_hurley-sunrise__full_mix.wav

Development Data


Download dev1.zip (external link) (22 MB) (former development data of SiSEC 2008 professionally produced music recordings (external link) task)
Download dev2.zip (external link) (36 MB) (newly proposed data)
Download dev2_full_mix.zip (external link) (75 MB) (full-length recordings for newly proposed data)

The data consist of stereo WAV audio files, that can be imported in Matlab using the wavread command. These files are named {dev1,dev2}__ [ <author> ] - [ <song> ]__[ <snip> ]__ {mix,full_mix,<track>}.wav, where <author> is the author name, <song> is the song name, <snip> is a shortcut for snip information, and <track> is the separated track name (e.g., "vocals", "bass", etc.).

The data include the following mixtures (snips and full-length recordings):

dev1
  • dev1__bearlin-roads__snip_85_99__mix.wav
  • dev1__tamy-que_pena_tanto_faz__snip_6_19__mix.wav
dev2
  • dev2__another_dreamer-the_ones_we_love__snip_69_94__mix.wav
  • dev2__fort_minor-remember_the_name__snip_54_78__mix.wav
  • dev2__ultimate_nz_tour__snip_43_61__mix.wav
dev2_full_mix
  • dev2__another_dreamer-the_ones_we_love__full_mix.wav
  • dev2__fort_minor-remember_the_name__full_mix.wav
  • dev2__ultimate_nz_tour__full_mix.wav

Separated tracks files (needed for evaluation in dev1 and dev2) are in the corresponding folders named {dev1,dev2}__[<author>]-[<song>]__tracks .

License


All audio files are distributed under the terms different licenses, as listed below for each recodring:

All the former test and development data (test1 and dev1) are from MTG MASS database (external link) by M. Nxx.

All the remixes of newly proposed data (dev2 and test2) are done by Michel Desnoues from Telecom ParisTech .


Tasks


The following should be taken in to account:
  • Note that only 20 seconds snips are asked to be separated, and not full-length recordings.
  • Some track names below have the following meaning:
    • "vocals" = "a sum of any singing including main vocal, back vocals and singing in the reverb"
    • "drums" = "a sum of any drums including bass drum, hi-hat, snare etc."
    • "bass" = "bass guitar only (i.e., not bass drum)"

Test Tasks


test1__tamy-que_pena_tanto_faz__snip__mix.wav
Extract the following stereo tracks:
  • vocals
  • guitar

test1__bearlin-roads__snip__mix.wav
Extract the following stereo tracks:
  • vocals
  • bass
  • piano

test2__glen_philips-the_spirit_of_shackleton__snip_163_185__mix.wav
Extract the following stereo tracks:
  • vocals
  • drums
  • bass

test2__nine_inch_nails-the_good_soldier__snip_104_125__mix.wav
Extract the following stereo tracks:
  • vocals
  • drums

test2__shannon_hurley-sunrise__snip_62_85__mix.wav
Extract the following stereo tracks:
  • vocals
  • drums
  • bass
  • piano

Development Tasks


dev2__another_dreamer-the_ones_we_love__snip_69_94__mix.wav
Extract the following stereo tracks:
  • vocals
  • drums
  • guitar

dev2__fort_minor-remember_the_name__snip_54_78__mix.wav
Extract the following stereo tracks:
  • vocals
  • drums
  • bass
  • claps

dev2__ultimate_nz_tour__snip_43_61__mix.wav
Extract the following stereo tracks:
  • vocals
  • drums
  • bass

Submission


Participants may submit separation results for any above-mentioned tracks of any above (test and development) 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 only 20 seconds snips are asked to be separated, and not full-length recordings.

Please submit your results through this link (external link). Please register the system first, then you'll receive your ID and password to login the system.

The submitted audio files will be made available on a website under the terms of the same license as indicated in the section Licenses above. In other words, any modified version inherit exactly the same license as the original.

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.

In addition, new auditory-motivated objective measures will be used to assess the quality of the estimated spatial source image signals, in the mono and stereo cases. The main features are:
  • as in previous evaluations, four performance measures akin to SDR, ISR, SIR and SAR are given: global score, target preservation score, interference rejection score and artifacts absence score
  • these auditory-motivated measures were trained on a set of subjective quality ratings obtained from the SISEC 2009 sound material and improve correlation to subjective measures by more than 20% compared to classical SDR, ISR, SIR and SAR
Code of the auditory-motivated measures is available here (external link).
  • A preliminary version of this toolbox is available here. It provides a new method to decompose the distortion into three components: target distortion "eTarget" (error signal related to the target source), interference eInterf (error signal related to the other sources) and artifacts eArtif (remaining error signal).

Potential participants


  • M. Nxx
  • Vasileios Pantazis
  • Alexey Ozerov (alexey.ozerov (a) irisa_fr)
  • Jeanlouis Durrieu (durrieu (a) enst_fr)
  • Maximo Cobos (mcobos (a) iteam_upv_es)
  • Pablo Cancela (pcancela (a) gmail.com)

Task proposed by Audio Committee and Olivier Le Blouch

Back to Audio source separation top

Menu