Informed Audio Source Separation: A Comparative Study

Antoine Liutkus (1) , Stanislaw Gorlow (2) , Nicolas Sturmel (3) , Shuhua Zhang (4) , Laurent Girin (4) , Roland Badeau (1) , Laurent Daudet (3) , Sylvain Marchand (5) , Gael Richard (1, 6)

1 LTCI - Laboratoire Traitement et Communication de l'Information
2 LaBRI - Laboratoire Bordelais de Recherche en Informatique
3 IL - Institut Langevin - Ondes et Images (UMR7587)
4 GIPSA-MAGIC - GIPSA - Machines parlantes, Gestes oro-faciaux, Interaction Face-à-face, Communication augmentée
5 Lab-STICC_UBO_CID_IHSEV
6 TSI - Département Traitement du Signal et des Images

The goal of source separation algorithms is to recover the constituent sources, or audio objects, from their mixture. However, blind algorithms still do not yield estimates of sufficient quality for many practical uses. Informed Source Separation (ISS) is a solution to make separation robust when the audio objects are known during a so-called encoding stage. During that stage, a small amount of side information is computed and transmitted with the mixture. At a decoding stage, when the sources are no longer available, the mixture is processed using the side information to recover the audio objects, thus greatly improving the quality of the estimates at a cost of additional bitrate which depends on the size of the side information. In this study, we compare six methods from the state of the art in terms of quality versus bitrate, and show that a good separation performance can be attained at competitive bitrates.

Domaines: Son [cs.SD]

Antoine Liutkus, Stanislaw Gorlow, Nicolas Sturmel, Shuhua Zhang, Laurent Girin, et al.. Informed Audio Source Separation: A Comparative Study. EUSIPCO 2012 - 20th European Signal Processing Conference, Aug 2012, Bucarest, Romania. pp.n/c. ⟨hal-00809525⟩