Abstract: This paper presents an algorithm to automatically analyze and extract soccer program highlights scene based on MPEG compressed audio-track analysis.In this algorithm,audio compressed features are directly extracted first and then based on the features the algorithm detects and segments commentary speeches,to recognize in particular the events of excited commentary and crowd cheers respectively.Finally,the recognized results are integrated by probability fusion,and the corresponding video clips of fusion result are chosen as soccer highlights.Experimental data shows that the algorithm works well.