ISCA Archive Interspeech 2006
ISCA Archive Interspeech 2006

Continual on-line monitoring of Czech spoken broadcast programs

Jan Nouza, Jindrich Zdansky, Petr Cerva, Jan Kolorenc

In the paper we describe the development of the first practical system that performs automatic on-line monitoring of Czech broadcast stations. It is based on our own speech recognition server that operates with 300K word lexicon and 2.3 RT factor. For true on-line service, several servers are connected to the platform that controls acoustic stream segmentation, distribution of data to the servers, collection of results and production of the final transcription. We show practical results achieved on different types of broadcast programs, such as news (21% WER), parliament debates (21% WER) and talk-shows (34%).