• MTG MASS database: MTG MASS
  To build our database, we used all the provided files from the MTG MASS database taht contained a solo voice (more specifically: a singer voice):
  
  • Bearlin: Roads, 85-99,
  • Fort Minor: Remember the Name, 127-145,
  • Fort Minor: Remember the Name, 199-209,
  • Kismet: TV on, 100-128,
  • Tamy: Que Pena Tanto Faz, 6-19,
  • Tamy: Que Pena Tanto Faz, 58-79,
  • Vieux Farka Touré: Ana, 30-55,
  • Vieux Farka Touré: Ana, 201-213,
    
  and then generated synthetic mixes such that each separate track (i.e. instrument) was "localized" in a static position.
  The ground truth for our solo/accompaniment task was then generated as 2 tracks: the first one contains the solo part, while the other one contains all the other sources.
• We also used the SiSEC08 database during the development of the algorithm, in order to participate to the evaluation campaign: SiSEC 2008 Professionally Produced Recordings.

• MTG Database (synthetic instantaneous + panning mixes):
  

  song	Original	SDR	After melody tracking...	SDR	... and with unvoiced model.	SDR
  Bearlin	solo voice	-3.4	est. solo (V-IMM)	7.2	est. solo (VU-IMM)	7.7
  	accompaniment	3.4	est. acc. (V-IMM)	10.6	est. acc. (VU-IMM)	11.1
  	mixture		Consonants more audible in solo extracted with VU-IMM
  Tamy	solo voice	2.6	est. solo (V-IMM)	11.4	est. solo (VU-IMM)	11.6
  	accompaniment	-2.6	est. acc. (V-IMM)	8.8	est. acc. (VU-IMM)	9.0
  	mixture		Consonants more audible in solo extracted with VU-IMM
  Vieux Farka Touré	solo	-8.6	est. solo (V-IMM)	7.8	est. solo (VU-IMM)	7.0
  	acc.	8.6	est. acc. (V-IMM)	16.4	est. acc. (VU-IMM)	15.6
  	mix.		Slight drop of performance for VU-IMM: some drum sounds taken in unvoiced parts of solo.
  Fort Minor	solo voice	-2.9	est. solo (V-IMM)	4.4	est. solo (VU-IMM)	4.8
  	accompaniment	2.9	est. acc. (V-IMM)	7.3	est. acc. (VU-IMM)	7.7
  	mixture		Rap song: melody hard to estimate.

   
• Comparing the mono algorithm [Durrieu09] with the proposed stereo
  

  Song	Original	SDR	[Durrieu09]	SDR	Proposed Algorithm	SDR
  Tamy	solo voice	1.7	est. solo	8.7	est. solo	12.1
  	accompaniment	-1.7	est. acc.	7.0	est. acc.	10.3
  	mixture		Special case: guitar panned to the left, singer to the right; spatially "easy", less favorable for mono algorithm.

  
  
  
• Samples from Take Five, by the Dave Brubeck Quartet, isolating Desmond from the rest:
  

  Song	Original mixture	After melody tracking...	... and with unvoiced model.
  Excerpt 1		estimated solo (V-IMM)	est. solo (VU-IMM)
  		estimated acc. (V-IMM)	est. acc. (VU-IMM)
  Excerpt 2		est. solo (V-IMM)	est. solo (VU-IMM)
  		est. acc. (V-IMM)	est. acc. (VU-IMM)

N.B.: Our estimated signals for the SiSEC08 evaluation campaign are also available on the result webpage of the event: SiSEC 2008 Professionally Produced Recordings Results (proposed algorithm as algorithm 3 for VU-IMM and algorithm 4 for V-IMM)

References:
[Durrieu09] J.-L. Durrieu, G. Richard and B. David, An Iterative Approach to Monaural Musical Mixture De-Soloing, ICASSP 2009. [pdf][poster][audio examples][copyright]

Copyright 2009 IEEE. Published in the IEEE 2009 International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2009), scheduled for April 19 - 24, 2009 in Taipei, Taiwan Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works, must be obtained from the IEEE. Contact: Manager, Copyrights and Permissions / IEEE Service Center / 445 Hoes Lane / P.O. Box 1331 / Piscataway, NJ 08855-1331, USA. Telephone: + Intl. 908-562-3966.