Carnegie Mellon University, Pittsburgh, has received a $4.8m four-year grant from the US National Science Foundation, the Advanced Research Projects Agency and NASA to develop an on-line, interactive digital video library system that will enhance the study of science and mathematics for students in kindergarten through to college. The Informedia Digital Video Library, to be created by Carnegie Mellon and public television station WQED/Pittsburgh, will integrate speech, image and natural language understanding technologies developed by university researchers to access, explore and retrieve video material from the archives of public television and educational institutions. Users will be able to search through the complete content of the stored video, retrieve segments of interest and view them on the screens of desktop computers over local networks. Initially, the library will contain about 1,000 hours of raw and edited video from WQED archives. Digital Equipment Corp will supply the project’s equipment infrastructure, including site servers and user desktops from personal computers to Alpha AXP-based workstations and video servers. Microsoft Corp will be contributing software technology and financial assistance and Bell Atlantic Corp will provide funding for the system’s communication services. The library system will be tested initially by Carnegie Mellon undergraduates; students at the Winchester Thurston School, an independent school in Pittsburgh; the Fairfax County schools and the Open University. Institutions outside Pittsburgh will access and retrieve library materials through a technique called pre-fetching via the Internet. A derivative of Carnegie Mellon’s Sphinx 2, a highly accurate, speaker-independent speech recogniser, will automatically transcribe video sound tracks, which will then be stored in a full-text information retrieval system developed at the university’s Center for Machine Translation. The system enables rapid retrieval of individual video sequences based on key words in the sound track. Users will be able to scan video rapidly without a loss of perception, as they might when skimming written material. Vision technology developed at Carnegie Mellon’s Robotics Institute can segment the videotape into meaningful units called video paragraphs. Researchers at the university’s Information Networking Institute will study issues of pricing and charging for use of the video material. They will implement a network-based billing system called NetBill that enables charging for small amounts of information transmitted over the networks. They also will incorporate mechanisms to ensure the privacy and security of people using the system and the material in it.
