Spoken information retrieval for multimedia databases

Summary form only given. This document describes the realization of a spoken information retrieval system and its application to word search into indexed multimedia databases. The multimedia database is build from a multiformat set of text, audio and video documents. The whole archive collection is indexed using preprocessing techniques to produce transcripts and indexing software tools to catalog them. The system uses a Java-based distributed client-server architecture. A Java applet is used to capture the audio signal for a spoken query, then it is transmitted to a server where an automatic speech recognition (ASR) software is applied to convert the signal into a transcripted hypothesis. Later, a query tool process the transcript sentence along with the indexed multimedia database and a set of pointers to documents are generated. Finally, a Web page with links to the resulting documents, where queried words appear, is presented to the user.