To retrieve interesting broadcast news documents out of an enormous number of TV news programs, if no indexing is done on
the news and word-based retrieval is required, it is inevitably necessary to transcribe all the broadcast news documents automatically
[1,2]. And this task can be done only by using a Large Vocabulary Continuous Speech Recognition (LVCSR) system.
In this study, first the retrieval performance was experimentally compared between the system using automatically transcribed
database and the one using manually transcribed database. Next, to input keywords via voice, we proposed a retrieval technique
using the association degree between keywords.