Research, Content-based Audio Retrieval Method

The research of audio retrieval based on content, as a newly arisen field, is still being researched and investigated at home and abroad. The audio frequency signal includes two types of signals ,speech and non-speech. Always, the processing of the audio frequency signal is mainly concentrated on speech processing, such as speech recognition and talks identify etc. The research of audio retrieval based on content is still not much. How to get the structure information and content meaning ,make audio frequency signal having the same semantic classes is the key of the research of audio retrieval based on content. Only there is breakthrough in audio frequency signal recognition based on physics characteristic, more and deeper research of sound index could be done.In this paper, we present a way of audio recognition based on MFCC and analyze MEL coefficient deeply. Through experiments, it proved that using average MEL coefficients as characters, DTW arithmetic is an efficient arithmetic for recognizing single sound signal.The major work and achievement of this paper are presented as follows: (1) we review the main methods of audio retrieval at home and abroad. Study the familiar audio frequency data processing technique and general methods of audio retrieval.(2)Study the character of audio signal. Analyze the zero-crossing rate and MFCC. A mean Mel coefficients is proposed, it can be used to recognized different audio signal.(3)Study the segmentation and recognition of audio frequency signal. Audio signal can be divided into segments based on zero-crossing rate. (4)A audio recognition arithmetic based on MFCC is proposed. Through this arithmetic, audio clip can be identified effectively.The simulation and data analysis in PC with VC6.0 software platform are carried out. And more simulation of audio segmentation and retrieval is done with MATLAB software platform.