Abstract
Now, It is difficult to watch a complete TV program for people. So, how to extract the most important or interesting fragments become a popular topic for researchers. In this paper, we used the characteristic of HMM to detect the audio events for the sports videos. We also used MFCC, Spectrum Centroid, Spectrum, Spectral Rolloff Spread to regard as the features of the audio events. We designed two methods in the experiment: 1. We try to find the best number of states for describing the audio events. 2. We try to find the best rang of Triangular band pass filters for each audio event. Finally, we compared the results for proving our method can be used in other sports videos.