Abstract
This paper describes a new video content skimming technique based on automatic extraction of emphasized speech and an evaluation on which the technique is applied to video mail contents. The technique analyzes speech information in original contents statistically and estimates emphasized portions of speech track, by using prosodic parameters such as pitch, power and speaking rate. To extract emphasized portions of speech, the method introduces a metric, “degree of emphasis" that indicates how much emphasized each speech segment is. When a user requests the video content of the article, the method collects the video segments corresponding emphasized speech potion referring to the user specified ' digest rate' and creates the new video sequence. In the evaluation, we used two different kinds of video contents to apply the proposed technique; one is non-edited video such as home video and video mail, the other is fully-edited video such as sports and dramas. The result of the experiments show that the sequence of segmented video have enough quality in general; however, the technique cannot be applied sufficiently to fully-edited video contents.
