Rozpoznávání klíčových slov a akcí v audiovizuálních datech

The goal of the project is automatic processing of recorded meeting data, which
would allow to user the direct browsing of required events in acoustic/visual
databases according to selected criteria. Further, proposed algorithms will be
also modified to be used in large vocabulary continuous speech recognition tasks.
Derivation of robust, efficient and by any user utilized system depends on
training and testing of particular algorithms. These algorithms are generally
very computationally expensive, mainly in case of processing real meeting data
and large vocabulary databases. A cluster proposed in this project is able to
significantly increase a computational capacity at our faculty, thus, new modern
algorithms and trends in area of automatic speech recognition can be applied. We
also suppose the utilization of new cluster in other spheres of research activity
at Faculty of Information Technology.