BSS_ORACLE

A toolbox to compute oracle estimators for source separation

by E. Vincent and R. Gribonval

Purpose

BSS_ORACLE is a MATLAB toolbox to evaluate the best performance achievable by a class of source separation algorithms in an evaluation framework where the true reference sources are known.

What it can do / what it cannot do

BSS_ORACLE can only compute oracle estimates when the true reference sources are known. In no way it provides a method to separate the sources blindly: it needs to know the reference sources to select the best separation algorithm (the oracle) within a given class.

Download (distributed under the terms of the GNU GENERAL PUBLIC LICENSE)

The most recent available version is version 1.0 [Matlab]. It implements oracles for
- stationary filtering (ICA-like algorithms)
- time-frequency masking (DUET-like algorithms)

Examples

Examples from our paper [1] below can be reproduced using the following data and matlab code.

References

E. Vincent and R. Gribonval, Construction d'estimateurs oracles pour la séparation de sources, to appear in Proc. GRETSI 2005, September 2005, Louvain la Neuve, Belgium.
E. Vincent and R. Gribonval, BSS_ORACLE Toolbox User Guide, Version 1.0, 2005, [Online http://www.irisa.fr/metiss/bss_oracle/]

Last modified: Mon Sep 05 10:12:22 +0200 2005