Manual labeling, as a conventional audio retrieval method cannot meet the demand of multimedia retrieval for its low efficiency with the explosive growth in the amount of multimedia information. Consequently, content-based audio retrieval is being paid more and more attention so as to improve the efficiency, effectiveness and accessibility of audio retrieval. In this paper, an approach based on general model template is proposed after the basic theory of general model and audio feature extraction based on independent component analysis are described. Corresponding algorithms are also detailed and experiment results presented to prove the effectiveness of this method.