SEGMENTING TV SERIES INTO SCENES USING SPEAKER DIARIZATION

In this paper, we propose a novel approach to perform scene segmentation of TV series. Using the output of our existing speaker diarization system, any temporal segment of the vid eo can be described as a binary feature vector. A straightforwa rd segmentation algorithm then allows to group similar contig uous speaker segments into scenes. An additional visual-onl y color-based segmentation is then used to refine the first segmentation. Experiments are performed on a subset of the Ally McBealTV series and show promising results, obtained with a rule-free and generic method. For comparison purposes, te st corpus annotations and description are made available to th e community.