Based On The Humming MIDI Music Retrieval System Research

In the continuous development of the technology and the network environment ofrapid growth in a number of multimedia technology and equipment, the audio data, asan important part of the multimedia data, the amount of information is constantlygrowing. In the massive music database, the traditional audio retrieval methods bringall sorts of restrictions to the use of the users, and users have not been satisfied with thetext-based audio retrieval, which proposed newer and higher requirements for musicretrieval. Natural, convenient and effective humming music retrieval is a content-basedmusic retrieval methods, which has a broad application prospects and importantresearch value. This mode allows the users to retrieve the desired songs by hummingform. The user does not need to remember the name of the song, artist or lyrics, and hejust hum the melody of the song to find his desired song.This paper researched the key technologies of MIDI music database’s hummingmusic retrieval system, mainly in three aspects: the MIDI music database melodicfeature extraction, the humming feature extraction and music melody matchingalgorithm. For these three aspects, this paper has done the following work:1. The MIDI format was chosen as the file format of music, the structure of theMIDI file was analyzed, and the melody information was extracted from the MIDImusic library.2. A Several pitch extraction algorithms were detailed analyzed. In order toensure the system has lower time complexity and higher algorithm accuracy.Thispaper used two layers BP neural to network to break musical note and distinguishdevoicing of humming voice signal. This paper also elaborated the average magnitude difference function and the advantages and disadvantages of the auto-correlationfunction method and further improved to obtain more accurate pitch period, so as toextract the pitch and duration, and generate the melody characteristics of humming,which was ready for subsequent matching algorithm.3. The advantages and disadvantages of several traditional melody matchingalgorithms were analyzed. Combined with the system requirements for accuracy andspeed, this paper constructed weighting sound long ratio which was based on the pitchdifference hidden Markov model to match, and ultimately got the best matching musicto match.Finally, the paper validated the MIDI music retrieval system based on humming,analyzed its noise immunity and compared with other papers. The results show that theretrieval of the system with high accuracy and practicality.