A Fuzzy Logic Approach for Content-Based Audio Classification and Boolean Retrieval

Since the invention of fuzzy sets and maturing of the fuzzy logic theory, fuzzy logic systems have been widely applied to various fields, such as fuzzy controller , data mining, and so on. New potential areas using fuzzy logic are also being explored with the emergence of other technologies. One booming technology today is the Internet, due to its fast growing number of users and rich contents. With huge data storage and speedy networks becoming available, multimedia contents like image , video, and audio are fast increasing. In order to search and index these media effectively, various content-based multimedia retrieval systems have been studied. In this chapter, we introduce a fuzzy logic approach for hierarchical content-based audio classification and boolean retrieval, which is intuitive due to the fuzzy nature of human perception of audio, especially audio clips of mixed types. The fuzzy nature of audio search lies in the facts that (1) both the query and target are approximations of the user's memory and desire and (2) exact matching is sometimes impossible or impractical. Therefore, fuzzy logic systems are a natural choice in audio classification and retrieval. The fuzzy tree classifier is the core of the hierarchical content-based audio classification. At the beginning, audio features are extracted for audio samples in the database. Proper features are then selected and used as input to a constructed fuzzy inference system (FIS). The outputs of the FIS are two types of hierarchical audio classes. The membership functions and rules are derived from the distributions of the audio features. Non-speech and music sounds are discriminated by the FIS in the first hierarchy. Secondly, music and speech are separated. One particular sound, the telephone ring, has also been recognized in this level. In the prototype system, the classification ability of up to fourth level has been explored. Hence we can use multiple FISs to form the 'fuzzy tree' for retrieval of different types of audio clips. With this approach, we can classify and retrieve generic audios using fewer features and less computation time, compared to other existing approaches. As for retrieval, the existing content-based audio retrieval systems usually adopt the query-by-example mechanism to search for desired audio files. However, only one single audio sample often cannot express the user's needs adequately. To overcome this problem, more audio files can be chosen as queries provided by the user or through feedback during searching. Correspondingly, we present …