Audiotory Movie Summarization by Detecting Scene Changes and Sound Events

A novel movie audio summarization framework is presented, which consists of three processing levels, namely, low-level audio feature extraction, mid-level audio event detection, and high-level auditory movie summarization. We first detect auditory changes in the eigen-audiospace to segment movie audio streams, then exploit a scoring algorithm to refine the segments. Audio events from movies are identified in a hierarchical manner from background detection, foreground event separation to key audio event identification, thus generating the final audio summaries from each movie. The experiments on different auditory categories from movies and TVs demonstrate the effectiveness of the propose approach.