Vast amounts of data exist in the form of audio/video tapes. In order for this data to be useful for digital libraries, it must be cataloged, indexed, and made easily searchable. In many cases the speech information contained in the audio track is of major interest. For example, a user might want to make a query of the form―find all instances where the words "balanced budget" and "tax cuts" are mentioned within 30 seconds of each other. Or a user may want to make a search based on both spoken materials and broad characteristics of the visual content. For example, a user make ask to find examples with the words "end of the cold war " are combined with visual clips of missiles. In order to make such queries feasible, a database must be properly labeled as to audio and video content, indexed, and integrated with appropriate database search tools. Software tools to prepare the audio/video content are now becoming available. In this paper, results are given as to the accuracy of the automatic speech recognition components of two of the leading software tools for creating such indexed digital libraries.
展开▼
