Audio-based description and structuring of videos

摘要

Enabling a rapid on-the-fly view of the content of a movie requires segmenting the movie and describing the segments in a user-compatible manner. The difficulty resides in extracting relevant semantic information from the audiovisual signal, both for the segmentation and the description. We introduce in this paper audio scenes and chapters in movies and present an algorithm for automatically segmenting a video based on the audio stream only. Audio scenes and chapters are defined as the equivalent of shots and scenes in the visual domain. A tree-like audio-based structure of a video is proposed. A chapter is then classified into different chapter categories. The automatic solution to audio scene and chapter segmentation and classification is evaluated on manually segmented and classified videos.