Research On Content-Based Audio Classification

Content-based audio classification is an important application direction of pattern recognition technology. Audio feature extraction and content-based audio classification are the technical bases of audio retrieval and video analysis. There are two key problems of content-based audio classification, which are how to extract more robust features from audio signal and how to improve the performance of audio classification models based on the current features.By expounding the development status of content-based audio classification nowadays, it can be shown that audio feature extraction and audio classification are two key techniques in the content-based audio classification system. The audio in soccer game, as an object researched, is divided into five classes, such as noise, commentator speech, whistle, cheers and commentator speech with background noise. According to the the MPEG-7 audio standard, the audio features are extracted at frame-level and clip-level. The audio perceptive features describe the content of the audio in soccer game and the different features combination can be applied in the audio classification. By deeply analyzing discriminating characteristics of audio information, the feature vector is formed and a hierarchy audio classifier based on decision tree is realized. Different from the traditional ways, the hierarchy audio classifier is not trying to apply one kind of rule, but to devide a complex classification into several simple ones and to adopt a method of step-by-step classification in order that the efficiency of audio classification can be improved greatly.The experiment results show that the features selected are effective and the classification accuracy is good. With some key technologies being solved, all of the work will be applied in practice greatly.