Improving Access to Digital Video Archives through Informedia Technology

摘要

Informedia research at Carnegie Mellon University combines speech recognition, image processing, and natural language processing to automatically index a digital video library. This engineering report focuses on the contribution of speech analysis for transcript generation And alignment, and the use of these features in library interface development. By deepening The automated analysis, such as using named entity extraction to identify people and place Names in the audio transcript, better summaries and visualizations can be produced to navigate Through video libraries holding thousands of hours of material.