BSS Eval

A toolbox for performance measurement in (blind) source separation

Purpose

BSS Eval is a MATLAB toolbox to measure the performance of (blind) source separation algorithms within an evaluation framework where the original source signals are available as ground truth [1, 3]. The measures are based on the decomposition of each estimated source signal into a number of contributions corresponding to the target source, interference from unwanted sources, and artifacts such as "musical noise". They are valid for any type of data (audio, biomedical, etc), any mixture (instantaneous, convolutive, etc) and any algorithm (beamforming, ICA, time-frequency masking, etc).

For audio data, the resulting energy ratio criteria correlate with subjective ratings to a certain extent only. For improved correlation with subjective ratings, try our latest toolkit PEASS.

Download (distributed under the terms of the GNU General Public License)

Version 3.0 for Matlab (18 kB) by E. Vincent
This version provides three complementary functions forming the core of the annual Signal Separation Evaluation Campaign [3]:
1. bss_eval_sources.m: for the evaluation of estimated single-channel source signals [1]
2. bss_eval_images.m: for the evaluation of estimated multichannel spatial source images [3, 4]
3. bss_eval_mix.m: for the evaluation of estimated mixing filters [3, 5]
Version 2.1 for Matlab (272 kB) by C. Févotte, R. Gribonval and E. Vincent
This version was developed with the support of GDR ISIS. It provides detailed evaluation capabilities (distinction between additive noise and other interfering sources, time-varying performance metrics) [1, 2] but is practically restricted to single-channel source signals within instantaneous mixtures. For convolutive mixtures, the use of version 3.0 is recommended due to a greater computational efficiency enabling longer decomposition filters hence better correlation with subjective ratings.

Examples

For examples of use of Version 3, see example_inst.m and example_conv.m on the Signal Separation Evaluation Campaign website.

Examples of use of Version 2.1 taken from our paper [1] are available here.

References

E. Vincent, C. Févotte and R. Gribonval, Performance measurement in blind audio source separation, IEEE Trans. Audio, Speech and Language Processing, 14(4), pp 1462-1469, 2006.
C. Févotte, R. Gribonval and E. Vincent, BSS_EVAL toolbox user guide - Revision 2.0, Technical Report 1706, IRISA, April 2005.
E. Vincent, S. Araki, F.J. Theis, G. Nolte, P. Bofill, H. Sawada, A. Ozerov, B.V. Gowreesunker, D. Lutter and N.Q.K. Duong, The Signal Separation Evaluation Campaign (2007-2010): Achievements and remaining challenges, Signal Processing, 92, pp. 1928-1936, 2012.
E. Vincent, H. Sawada, P. Bofill, S. Makino and J.P. Rosca, First stereo audio source separation evaluation campaign: Data, algorithms and results, in Proc. 7th Int. Conf. on Independent Component Analysis and Signal Separation (ICA), pp 552-559, 2007.
E. Vincent, S. Araki and P. Bofill, The 2008 Signal Separation Evaluation Campaign: A community-based approach to large-scale evaluation, in Proc. 8th Int. Conf. on Independent Component Analysis and Signal Separation (ICA), pp 734-741, 2009.