SBIR-STTR Award

This Small Business Innovation Research Phase I project will demonstrate the feasibility of creating a system that automatically generates personalized broadcasts from a library of audio / video (rich media) content. Such a system is needed because individuals are overloaded with rich media content and lack advanced tools for navigating this deluge of rich media. Building upon StreamSage's existing expertise in dealing with rich media information, StreamSage's research will alleviate this problem by creating a system capable of creating "personalized rich media broadcasts" that automatically characterize the gaps between disjointed segments of content and fill these gaps with bridging text that provides necessary background and structure to the segments from multiple rich media files. The effort in Phase I will expand the current-state-of-the-art by developing algorithms capable of automatically identifying the types of gaps between the rich media segments and by establishing methods by which the information necessary for coherently bridging these gaps can be automatically extracted from the rich media files. The personalized broadcast system created by this research would greatly improve end-user interactions with the rich media content by intelligently ordering and bridging the content pushed or pulled to the end-user. Additionally, an automated personalized broadcast system would enable the pushing of rich media content on a large scale, which has been impossible to date because of the tremendous manual intervention required to create a broadcast of usable quality

This Small Business Innovation Research Phase II research project will create a prototype system that will cut through the overload of audio/video (rich media) content by generating personalized broadcasts from a library of rich media documents. Building upon existing expertise in dealing with rich media, the proposed research will apply and refine the techniques discovered in phase I to organize relevant material using both the context of the documents and the topics of the selected material. The prototype will also apply the phase I results to identify and fill in the critical gaps between segments of material extracted from the source documents with bridging text that will provide necessary context and structure, allowing the system to present the relevant material as a single coherent broadcast. This research will result in new techniques that allow separately obtained passages of audio/video (or even text) to be joined together coherently. It will also provide techniques for organizing information based on both contextual and topical cues. These techniques will be applicable in any context in which information in natural language form is being extracted from a source collection. Furthermore, the research results will provide cost efficiencies for a number of specific important vertical markets (e.g. finance, broadcast news monitoring, etc.). The resulting software products will dramatically reduce the costs of the currently manually intensive information extraction process employed by firms in these markets. More generally, the software products that are derived from the company's current technology platform will also increase individuals' ability to find and absorb relevant information from diverse information sources, many of which are entirely intractable today. This ability is important in a wide range of communities such as academic institutions, intelligence agencies, homeland security agencies, financial institutions, and news broadcasters.