A Novel Approach for Video Song Detection using Audio Clues from Bollywood Movies

This paper proposes a novel approach for detection of video songs sequences using audio clues from bollywood movies. Approach exploits the fact that how audio clues help us to differentiate song portion and non-song portion. The Proposed method relies on following phases: Movie segmentation, Audio Intensity Based Classifier (AIBC), Song Sequence Generator (SSG), and Audio Clues Based Classifier (ACBC). Our approach exploits the fact that songs sequences have continuity rather than non-songs sequences. Initially movie is divided into various segments using movie segmentation method. Song Sequence Generator (SSG) extracts promising song sequences which are used as input to the Audio Clues Based Classifier (ACBC).The method relies on audio clues namely Tempo and Silence ratio which leads to desired song sequences. The proposed method is tested on movies of various genres that cover Romantic, Comedy, Action, Suspense and Family and more than 90% accuracy is achieved.