Music Retrieval Methods For Query-by-Humming Information And Its Web Application System Implementation

Abstract:With the rapid development of the digital music, internet and mobile phone technology, it has become a trend that integrating the three technologies together to meet the demands of human daily life, learning and entertainment. In traditional retrival, the music is usually labeled by text which requires users to provide some keywords, such as the song’s name, author, lyrics and so on. With the accelerating pace of life, people are not satisfied with the text-based music retrieval. It is hoped that through humming a melodic residual fragment of the user’s mind as a searching input, the target song that contains the melodic fragment can be retrived. This kind of music retrival technology will bring great convenience to the users. Humming-based song retrival as a branch of the content-based music retrival, is an important research topic in the field of digital music. It has great significance in theoretical and practical applications. In the field of pattern recognition, music retrieval algorithms get more and more attention due to their simplicity and high efficiency, while the feature extraction and matching of melodic fragment is an difficulty in them.This paper focus on humming-based music retrival, the main results that are obtained are as follows:(1) As effective information contained in the MIDI music file are only the pitch and duration, we selected MIDI music file as a database to store music. Because the wav music files can save the collection and quantization of the melody fragment data, we selected wav music file as the humming music to store, and designed the two-dimensional melody of pitch and duration to match the two different music melody.(2) We use the normalized、band-pass filter、pre-emphasis、 window-frame、filtering the mute section to pretreatment the melodic fragment, use the autocorrelation function to extract the fundamental frequency, and use the median smoothing processing to filter noise point. Finally, the humming music fragment’s feature can be extracted. (3) This paper proposes a matching method of two approximate dynamic time warping based on the melody, we solved the time point note sequences alignment and relative pitch problems of the humming music retrieval, compared to the dynamic time warping algorithm, the music retrieval method of top ten hit rating increased by18.63%.(4) A web oriented humming-based song retrival system is designed and implemented. The system is linear, real-time and concurrent. Twenty-five Figure, ten tables, sixty-six references.