Video segmentation by audio analysis

In this paper, a novel approach based on acoustic cues for automatic segmenting television stream into individual programs is proposed. This presented method is composed of the following steps: Several sets of repetitions in the audio track is detected by using silence detection and robust audio hashing; The found repetitions are treated as advertisements if the range of their length is from 5 seconds to 120 seconds; Programs are segmented from the recorded TV streams using the detected advertisements. Experiments on real-world TV recordings show the effectiveness of the proposed approach.