Efficient general genre video abstraction scheme for embedded devices using pure audio cues

In this paper, we propose a framework of general genre (e.g., action, comedy, drama, documentary, musical, etc...) movie video abstraction scheme for embedded devices based on pure audio. The proposed algorithm does chaptering of multi-genre movie videos by detecting silence, environmental noise, pure speech, music (pure instrumental music and music with vocals, i.e., songs), and speech with back ground music (or music without vocals but with speech). Various audio features along with supervised classification strategies have been used for the abstraction. The current system has been evaluated with Gaussian Mixture Model (GMM) and Fuzzy Decision Tree (FDT) classifiers. The silence and environmental noise have been detected using the threshold approach with certain combination of audio features. Various optimizations done at algorithm and program level have made the scheme highly suitable for embedded devices.