Kvalitativní posun v automatickém rozpoznávání jazyků s využitím streamovaných audio-médií

The projects aims at massive usage of streamed audio for a qualitative
improvement of LID (automatic language identification) system accuracy. The
speech processing research group at Faculty of Information Technology, Brno
University of Technology (Speech@FIT) disposes of a state-of-the-art LID system
based on acoustic and phonotactic modeling. For further improvement of its
accuracy, it is crucial to gather huge amounts of language-specific data. In the
framework of this project, such data will be collected from available streamed
sources (Internet radios), on-line stored, parameterized and processed. We will
develop software for training of LID models. Resulting models and algorithms will
be evaluated in international evaluation campaigns organized by NIST and in
cooperation with Czech law enforcement forces.