Audio Humming Recognition And Searching System Based Content

In the era of digital, digital video, music, and cartoons, has entered our lives. The multimedia content, e.g. songs, is stored in databases indexed by their titles, authors, singers, etc. However, the rhythm of the song is remembered more easily than its title, author or singer. As the volume of multimedia database grows, it's impossible for people to remember the index of data (name, author, and so on) completely. Therefore, content retrieve has become more and more useful than text retrieve.In this paper, the development and related theoretical research work of humming identification and recognition system on digital audio is introduced. Several key technologies in the singing recognition are discussed in detail. On the DE2 and PC platforms, a simple humming recognition system which is used in demonstration is realized.Throughout the development process, the system is realized on two platforms: PC platform and DE2 platform. First humming identification based on 20 songs on the PC and DE2 platforms is realized. With sufficient experiments and parameter adjustment, several problems, such as feature extraction, noise removal, Eigenvalue normalization, Eigenvalue recognition, had been settled. Finally we got a high recognition rate and a better running time on DE2 board. Then the main research was transferred to the PC platform. An effective humming recognition and searching system based on a library of more than 30 songs was set up, meanwhile, the pitch normalization algorithm, improved DTW, and the partial matching algorithm, was studied. Based on the priority condition"near the beginning or end", the author creatively put forward a new partial matching algorithm, with the use of pros and cons DTW. The time complexity, effectiveness, and compatibility of this algorithm is deeply analized and researched, finally a more satisfactory result was got. On PC platform, with a library of 52 music section, an 85% success rate was got in the use of partial matching algorithm, compared with 48% without partial matching algorithm. The pros and cons DTW algorithm didn't introduce too much loss in time complexity, the overall running time is only about 1.5 times of the overall matching algorithm. While this algorithm retained the outstanding compatibility of overall algorithm, meet the actual requirements completely.