The Research And Implementation Of Singing Voice Detection System Based On Active Learning Technique

As the quick development of networking technique,the multimedia data was increasing heavily.The progress of audio compression technique and advent of the storage devices with large capacity enable the booming of audio data in the internet.All these information was extensively applied in the field of education,entertainment,news,advertise and so on.And all these have become an important part of human life.Therefore, how to effectively deal with the analysis,storage and retrieval of mass audio data has become an urgent problem.Compared with matured image and video retrieval technique,the audio retrieval technique was comparatively inferior.Recently,many studies raised many tasks on audio information retrieval, and most of them were based on so called content based audio retrieval which mainly adopted the construction of index and classification of audio information.But in here,we mainly focus on similarity based searching.It’s about the searching solely based on signal similarity in the set of unlabeled audio clips.For similarity based audio retrieval,a traditional way was based on the dependence of data itself or feature vectors based signal detection technique which are extracted from data.However,on the assumption of employing this technique,the computation cost for long time audio signal or many other refereed audio signal, the retrieval will be impractical. Certainly,we could accelerate the process by the rough matching,but still it will inevitably reduce the correctness and stability of retrieval.Consequently, we employed a new searching algorithm based on similarity, this algorithm offers faster search and maintains enough accuracy.The key to acceleration was an effective utilization of pruning algorithm of feature information.This pruning algorithm could also be called dynamical version of quick histogram search method.This paper implement a prototype system based on similarity, and achieved rapid efficient and accurate audio retrieval.Finally,we envision the trend of development of similarity based audio retrieval system.