Indexing of multilingual news telecast using audio-visual keywords

Indexing of news video streams with semantic keywords is of interest to agencies that regularly monitor many news channels. In this paper, we describe a new method for indexing news video in different languages, for which there are inadequate language tools. Our approach involves combining multimodal inputs, namely audio and visual, and spotting of a handful of keywords with higher reliability, as compared to creating complete transcripts. We conduct a set of experiments to establish performance improvement despite lesser reliability of the language tools.