Project : metiss
Section: Other Grants and Activities
The ELISA Consortium
The ELISA consortium was set up as a spontaneous non-funded initiative in 1997 by ENST, EPFL, IDIAP, IRISA and LIA.
Its objective is the development, maintenance and improvement of a speaker verification platform that is shared between the members of the Consortium and which is presented in the context of the NIST yearly evaluation in speaker recogntion and tracking.
In 2004, METISS has been participating for the 7th consecutive year to the NIST evaluation, with a system based on the ELISA platform, and obtained well-positioned performances. .
Since this year, a version of the ELISA platform is being consolidated in the context of the Technolangues AGILE project (ALIZE sub-package).
HASSIP Research Training Network
The HASSIP (Harmonic Analysis, Statistics in Signal and Image Processing) Research Training Network is a European network funded by the European Commission within the framework programme Improving the Human Potential. It started on October 1st 2002, with founding partners: Université de Provence/CNRS, University of Vienna, Cambridge University, Université Catholique de Louvain, EPFL, University of Bremen, University of Munich and Technion Institute.
One of the aims of the HASSIP network is to shorten the development cycle for new algorithms by bringing together those who are involved in this process: the mathematicians and physicists working on the foundations (with view towards applications), the partners doing applied research (mostly engineering departments), are more experienced when it comes to implementations. The main research goal is therefore to improve the link between the foundations and real word applications, by developing new nonstandard algorithms, by studying their behaviour on concrete tasks, and to look for innovative ways to circumvent shortcomings or satisfy additional request arising from the applications.
The main contributions of the METISS project-team at IRISA will consist in new statistical models of audio signals for coding and source separation, as well as theoretical contributions on time-frequency/time-scale analysis and (highly) nonlinear approximation with redundant dictionaries.