Based Humming Music Retrieval Technology

With the rapid development of multimedia and Internet techniques, the digital information of video and audio has taken the place of the simulated video and audio, and been found in the Internet everywhere. Recently, it is a hot problem how to retrieve the information that our need in large-scale digital audio information. Classical music retrieval method is based these attributes of music name, author and player. But this retrieval method is not ideal. In the view of information essence, this information is not music content but attached feature. Usually, human remember music by the music melody, so we retrieve music by humming the music melody. Recently, content-based music retrieval technique has been one of hot research problems, and it expresses, recognizes the music content by extracting pitch features form audio frequency, and retrieves music according to comparability measure. The two most important problems are extracting music feature and matching, retrieving Malcolm.By the music retrieval for background, this paper mainly researches the error when hamming, the extracting and retrieval algorithm of pitch, then on the basis of these, it designs a extracting method of pitch. Moreover, it presents a kind of wave ridge matching algorithm in the view of reducing hamming errors and the effect that no-precise extracting to retrieval. The main contributions of this paper are:(1) Profound research music knowledge with related to retrieval. It profound researches the four factors of music, making tonality rule of music, the relation of semitone and sound frequency. On the basis of these, it gets the difference of tonality between hamming and standard tonality. These work determinate extracting and quantification of feature, and how to reduce the effect that error hamming to the retrieval algorithm.(2) Research the application of digital signal in the process of extracting feature. It profound analyzes, researches the time-domain and frequency-domain. According to these problems that are human sound mingled in the music, various kinds of sound of musical instrument, applause and environment noise, it designs a kind of algorithm of pitch of main melody under the environment of complex sound. In the duration's time of sound aspect, it designs a kind of note division method that has no limit in hamming ways. This method is suitable to pitch feature extracted from the usual music. Because of these two algorithms, the system that is designed in this paper can use wav music in the music database. However most people use the midi music.(3) Present peak value alignment retrieval. This algorithm is different with (U, D, R), N-gram that are character string fuzzy match algorithms. It is different with based-HMM retrieval algorithm that is belongs to statistic model, it is belongs to outline comparing algorithm. This algorithm can easily reduce the effect that many kinds of errors of hamming to the procedure of retrieval. The speed of algorithm is better than (U, D, R) algorithm that representative part character string fuzzy match algorithm, and is greatly better than based-HMM model statistic retrieval algorithm. Experimental results show that the algorithm is effective, and in most cases users can retrieve object music in the front of theretrieval results.This paper proved the efficient of algorithms presented in the paper by the above work, and designed a prototype system that is hamming-based music retrieval. This paper has beneficial researched and explored in the content-based music retrieval aspect.