Using Improved Artificial Bee Colony Algorithm To Optimize The System Of Query By Humming Content-based Music Retrieval

With the development and the growing popularity of computer networks and multimedia technology, the multimedia information needs are more and more bigger. As an important member of the multimedia information, digital music is also more frequently affecting people's daily lives. Music retrieval gradually become a new popular research field. How convenient, natural, quickly and accurately find the desired music is becoming one of the hot technology of the next generation of search engine. Query by humming content-based can help users find the music by humming the melody clips. The novel music retrieval method has great practical significance to the mobile phone network and the Internet.Query by humming content-based music retrieval mainly related to the problems of music melody feature extraction, music melody matching and music database construction. Based on the existing research results, we built a humming retrieval system by using the melodic contour as music feature. In order to improve the search accuracy of the system, we introduced an optimization technology to optimize some key parameters. The main work and the research contribution are as follows:(1). Improve the Artifical Bee Colony optimization: Based on Artifical Bee Colony algorithm, we divided the population into several independent sub-populations, then executed the various branches with each sub-population by artificial bee colony algorithm in parallel. After that, the branches communicated with each other by two different strategies. Three well-known benchmark functions were used to assess the performance of the improved artificial bee colony algorithm and the test results were compared with the original algorithm. In the first improvement, The accuracy was about 53% higher than the orginal algorithm and the convergence rate was about 9% faster than the orginal one. In the second improvement, The accuracy was about 73% higher than the orginal algorithm and the convergence rate was about 4% faster than the orginal one.(2). Music retrieval system: First, introduced the basic framework of Query byhumming content-based music retrieval system (QHCMR). Then described each function module of the system in detail and completed them. After that, we defined the evaluation criteria of the matching process (relevance) and given the whole system (accuracy). Finally, we tested QHCMR system.(3). Parameter optimization: We used the improved artificial bee colony algorithm to optimized some key parameters of the QHCMR system. Then we used the optimized parameters to test the QHCMR system and compared the test result with the orginal system. The experimental results showed that using the optimization method to optimize the parameters of QHCMR system would be useful and the performance of the optimized system would improve.