Implementing An Audio Clip Retrieval System With Java

Audio retrieval is one of the most important means to obtain information.How to get the content from the massive audio clips quickly and accurately becomes the main bottleneck in the development of audio retrieval.The use of audio fingerprinting technology can effectively solve the problem of the storage of huge audio data.However,it can only fuzzy match the similar audio.Therefore,the paper focuses on how to select audio features,design retrieval methods based on a method combined by Hierarchical SVM classification and fuzzy clustering,set up an audio retrieval system and test the performance of the system,in order to solve the contradiction between audio retrieval precision and retrieval speed in the case of large amount of audio data,The paper uses Java language to design and implement a prototype system of audio clip retrieving based on feature similarity.The system realizes high speed and effective audio clip retrieval and it has good expansibility.The prototype system is an experimental platform,and the performance test verified its usability and reliability.In the paper,we first discuss the structure of audio feature vector.Secondly,it designs a modified hierarchical SVM classification method by comparing the commonly used audio classification methods,which can improve the classification accuracy to more than 90%.In addition,it can maintain the classification accuracy of unknown data in practical application.The purpose of using audio classification technology is to reduce the search range,reduce the number of audio comparison,and reduce the retrieval time.After that,the paper discusses the method for calculating the similarity of audio clips,and implements an audio segment retrieval system based on fuzzy clustering.The paper also carries out many experiments to verify the performance of the system.Finally,according to the final test results,the paper prospects the development of audio retrieval system.