SiSEC 2013 - 4th Community-Based Signal Separation Evaluation Campaign (SiSEC 2013)

Following the success of the three previous editions, we are glad to
announce the launch of the SiSEC 2013 campaign for the evaluation of
source separation algorithms. This campaign is open to anyone, whether
you are an expert or a newcomer, you are interested in e.g. audio,
biomedical signal or image processing, and your algorithm relies on prior
information or not.
As a first step, we are issuing a call for task proposals. The goal is to
raise new separation tasks that are related to hot LVA/ICA topics (especially
blind and semi-blind separation, NMF, tensor decomposition, sparse component
analysis, approximate joint-block diagonalization etc.) and could be of
interest to the LVA/ICA community.
This is a community-based endeavor: all aspects of evaluation,
including datasets and evaluation criteria, are to be agreed between the
participants. We organizers will address the time-consuming issues of
collecting and evaluating the results submitted by all participants.
If you wish to propose a new task, contact the Chairs of SiSEC2013,
Nobutaka ONO ,
Zbynek Koldovsky , or
Toshihisa Tanaka
Three audio tasks are proposed as a starting point:
- underdetermined speech and music mixtures
- two-channel mixtures of speech and real-world background noise
- professionally produced music recordings
If you wish to comment upon the proposed tasks,
you can also post your comments on the wiki http://sisec.wiki.irisa.fr/
(register at http://sisec.wiki.irisa.fr/tiki-register.php).