| With the development of multimedia technology, multimedia information has been widely used in the area of information processing. Audio information play an important role in this area. Audio information are non-semantic notation and unstructured binary data stream which has caused great difficulties in audio signal processing and analysis. How to extract the audio semantic information is the key issue of the audio signal processing and content-based audio information retrieval. Audio classification technology is the major technology to solve this problem and it is the basis of audio content extraction. This thesis mainly focuses on how to select and extract audio features which can accurately describe the difference between various types of audio signals and how to make an effective audio signal classifier.In this paper, we study the existing audio classification methods. We design and implement a SVM-based audio classification module, and apply to many multimedia application systems. First, we analyze the existing short-term audio signal processing approach. We also study the methods of audio feature extraction and construct a set of feature vectors based on PCA-LDA. Then, we design and implement an audio classification module based on SVM. Finally, the audio classification module is embedded into an existing automatic subtitle generation system and pornographic information filtering system. The experiments show that using PCA-LDA method in audio feature extraction is effective for improving the audio classification results. |
