Multimedia Indexing
Pedro Moreno, Hewlett-Packard
During the last two years HP Cambridge Research Lab has explored indexing and search technologies and its application to several mulitimedia types such as audio, images and music. I'll describe some of the systems we have built with a special emphasis on our SpeechBot (http://www.speechbot.com) audio-indexing system. I'll give an overview of this audio search engine, its current limitations and some of the technologies we have explored to improve and extend its capabilities. Among others, I'll describe our experiements with Bayesian belief networks for topic segmentation, boosting for confidence scoring and particle (subword) based recognition for improved IR performance.
