Research On Audio Technology Of Query By Example Based On Shazam Algorithm

With the advent of the big data era,the amount of multimedia information increases explosively on the Internet.The traditional retrieval method based on text annotation can no longer meet people's demand for multimedia retrieval.The retrieval method based on the content information of multimedia files has become a hot research topic in recent years.Among them,the query-by-example(QbE)has the characteristics of convenient use,no need of labeling information.Taking audio as the example,people sumbit an unknown audio fragment first,and search the database to obtain relevant information of the audio.The Shazam audio retrieval is an important form of QbE.It has the advantages of small size and fast retrieval speed.This paper aims to improve the efficiency of user search by improving the traditional Shazam audio retrieval algorithm.The main work is as follows:1.Establish a baseline audio retrieval system.Shazam,a music retrieval algorithm,is introduced into the speech based QbE.And an audio based QbE baseline system is build.the performance of the system is verified on the self-built dataset finally.2.System optimization.In the retrieval and matching stage,the candidate audios,sorted by later,are removed firstly,and then takes the top N audios to find the maximum time offset,thus the retrieval time of the system can be reduced.The method of exchange efficiency based on index space,explores the influence of important parameters in the index space and chooses the most suitable parameters to construct the index space,which improved the system performance effectively.3.A new feature extraction algorithm is proposed.The feature extraction in the original Shazam algorithm selects the point of peak energy of each frame as the feature point,but the new feature extraction algorithm selects the energy threshold point based on the rectangular region as the feature point.First of all,this thesis analyzes the feasibility of the improved algorithm in theory.Afterwards,compared the two algorithms through the pure audio and another kind audio recorded in indoor noise environment respectively.It verifys that the improved algorithm can extract more representative feature points and the system's retrieve error rate is reduced greatly.Taking an 8s audio fragment as an example,the error rate of the improved algorithm is 55.3% lower than that of the original Shazam algorithm.4.Using C,Java programming language,and C/S mode,to achieve an audio based on QbE system on the APP client and server mode.A series of analysis of the design and implementation of the system are conducted in detailed,the main modules are introduced,and the system performance is tested as well.