Research On The Pitch-based Music Retrieval By Humming

With the rapid development of science and technology and the popularization of digital equipment, The number of multimedia is growing very fast. Audio data is one of an important part of multimedia data,Its information is expanding constantly.The traditional text-based audio retrieval can not meet the needs of the people for music retrieval. So finding a more rapid and effective information retrieval method to present has become an important task.The content-based music retrieval method is a new angle to distinguish music information, Such as amplitude, frequency, rhythm, pitch and other internal information. The hum-based music retrieval method’s core technology is the features extraction of humming music melody matching. This paper used the pitch and duration of the feature information to express the melody.The hum-based music retrieval method has a good application prospect for music’s quick inquiry, the singer’s training and helping the composer to remember music score, etc. It may be the next generation of music retrieval of the core technical one.This paper mainly content as follows.Firstly, This part introduced the basic signal pretreatment methods, including the wavelet de-noising, pre-weighting and adding window etc.; Secondly, the paper described the basic characteristics of speech signal in Time domain and frequency domain, including short-time energy, autocorrelation and zero-crossing rate, and then the paper proposed a method, which combined the wavelet analysis and linear prediction with the Autocorrelation and Average magnitude difference function. The result of experiment validated that this method could extract the pitch frequency precisely; Thirdly, This paper introduce and researches the traditional system retrieval matching algorithms. According to the note of pitch and duration, the article proposed a new algorithm, which combined a Hidden Markov Model based on the pitch difference with ratio of duration, Eventually, It got the best matching music; Finally, the paper research and design content-based humming music retrieval system, Experimental results show the system for music retrieval had a high accuracy and a relatively good robustness for all sorts of noise.