Content-Based Retrieval of Audio in News Broadcasts

Dogan E., SERT M., Yazici A.

8th International Conference on Flexible Query Answering Systems, Roskilde, Denmark, 26 - 28 October 2009, vol.5822, pp.548-559 identifier

  • Publication Type: Conference Paper / Full Text
  • Volume: 5822
  • City: Roskilde
  • Country: Denmark
  • Page Numbers: pp.548-559
  • Middle East Technical University Affiliated: Yes


This paper describes a complete, scalable and extensible content-based retrieval system for news broadcasts. Depending on segmentation results of the selected audio data, our system allows users to query audio data semantically by using both domain based fuzzy classes (anchor, commercial, reporter, sports, transition, weatherforecast, and venuesound) and similarity search. Two kinds of experiments were conducted on audio tracks of TRECVID news broadcasts to evaluate performance of the proposed query-by-example technique. The results obtained from our experiments demonstrate that Audio Spectrum Flatness feature in MPEG-7 standard performs better in music audio samples compared to other kinds of audio samples and the system is robust under different conditions.