Fast Caption Alignment for Automatic Indexing of Audio

For large archives of audio media, just as with text archives, indexing is important for allowing quick and accurate searches. Similar to text archives, audio archives can use text for indexing. Generating this text requires using transcripts of the spoken portions of the audio. From them, an alignment can be made that allows users to search for specific content and immediately view the content at the position where the search terms were spoken. Although previous research has addressed this issue, the solutions align the transcripts only in real-time or greater. In this paper, the authors propose AutoCAp. It is capable of producing accurate audio indexes in faster than real-time for archived audio and in real-time for live audio. In most cases it takes less than one quarter the original duration for archived audio. This paper discusses the architecture and evaluation of the AutoCAp project as well as two of its applications.