Narrative Peak Detection in Short-Form Documentaries Using Speech Recognition Transcripts

Narrative peaks are points at which the viewer perceives a spike in the level of dramatic tension within the narrative flow of a video. In this paper we describe two approaches for automatic identification of narrative peaks in short-form documentaries, within the framework of the VideoCLEF 2009 Affect Detection task. Both approaches exploit the speech recognition transcript in order to identify the narrative peaks. Our first approach is based on the idea that speech rate increases with high arousal and intensity and the second approach on the idea that narrative peaks are perceived where particular lexical items are used. These approaches are compared to a challenging baseline that predicts the presence of narrative peaks at fixed points in the video, presumed to be dictated by natural narrative rhythm or production convention. The second approach easily outperforms the challenge baseline while the first approach failed to achieve the performance level of the random baseline detector.