- Problem Definition
- The Solution
- Informedia Goals
- Speech Recognition in Informedia
- Conclusion
Problem Definition Currently, the TV and radio news is broadcast at a particular time, and if a person is not in front of a TV or radio at that time, the information becomes virtually inaccessible. There is simply not enough time to scan through tapes of yesterday's news for relevant stories. In addition, the viewer/listener must watch or listen to all stories in a news show, without the ability to select which stories to skip and which stories to pursue in more detail. Furthermore, a person can only attend to one news channel at a time. Similar or related information broadcast on another news channel at the same time cannot be viewed. The Informedia Solution The Informedia solution is to compress and digitally store news broadcasts on computer. At the user's convenience and within minutes of the conclusion of the program, all information is accessible through interactive queries. These queries allow the user to retrieve stories of interest from all the networks that carried stories on the topic of interest. An overview of the structure of the Informedia system is shown in Figure 1. While some Informedia educational testbed prototypes develop their library with computer-assisted methods that require human post-processing the News-on-Demand system is designed to be fully automatic. During the demonstration a user will speak queries about recent news events into the system. News-on-Demand will display icons for relevant stories, allowing the user to select interesting ones for playback. Informedia Goals The Informedia Digital Video Library Project at Carnegie Mellon University is creating a digital library of text, images, videos and audio data available for full content retrieval. Through the integration of technologies from the fields of natural language understanding, image processing, speech recognition and video compression, the Informedia System allows a user to explore multi-media data in depth as well as in breadth. The Informedia system for video libraries goes far beyond the current paradigm of video-on-demand, retrieving and displaying short video paragraphs in response to the user's query. As a result, a large body of video material can be searched with very little effort. While our work is centered around processing news stories from TV broadcasts, the system exemplifies an approach that can make any video, audio or text data accessible. The same methods can be used to index and search other streamed multi-media data by content. Speech Recognition in Informedia We can distinguish two distinct phases in the Informedia News-On-Demand process: library creation and library exploration. Library creation deals with the accumulation of information, transcription, segmentation and indexing. During library creation, speech recognition helps create time-aligned transcript of spoken words as well as correcting and integrating close-captioned text (if available). Library exploration concerns the interaction with the user trying to retrieve selections in the database. During library exploration, the CMU Sphinx speech recognizer allows a user to query the system by voice, simplifying the interface by making the interaction easier, more direct and more immediate. Conclusion The benefits of Informedia News-on-Demand are very dramatic. We can finally navigate the complex information space of news stories, without the linear access constraint that normally makes this process so time consuming. Thus Informedia News-on-Demand provides a new dimension in information access to video and audio material. In the future, we plan to add OCR capabilities for reading headlines and image processing for visual scene segmentation to the News-on-Demand system. |
