Research On The Content-based Humming Music Retrieval System

In the era of rapid development of computer multimedia technology,the number of songs in music database has increased dramatically.How to quickly and accurately find the required music data from the huge music database has become a hot research topic in the field of modern information retrieval.Traditional text-based music retrieval systems rely on the input of song name,singer name or song-related text information,which can not meet the actual needs of people for music retrieval.Therefore,content-based humming music retrieval system is the main research direction at present.This paper designs a humming music retrieval system based on DTW(dynamic time warping)algorithm.The system mainly includes three parts:preprocessing module,melody expression module and matching module.For the humming music data and the music database data,this paper uses two different methods to extract features.This paper proposes a two-level note segmentation method based on double-threshold endpoint detection and fundamental frequency curve fluctuation measure.The correct ratio of note segmentation is 90%,and the pitch difference and pitch-length ratio are used to express the relationship between the music signal notes of pitch and rhythm,so as to express the music melody.The HPSS(Harmonic Instrument/Percussion Instrument Sound Separation)method is used to filter and pretreat them.The harmonic component is separated from the mixed music signal,leaving only the mixed signal of the human voice and the percussion component.A main melody extraction algorithm based on auditory scene analysis is proposed to be used to extract the main melody of human voice.The DTW algorithm is optimized in matching module,which effectively improves the retrieval efficiency.The similarity of humming input is obtained by considering the combination of pitch similarity and rhythm similarity,which is used as the evaluation standard of humming music retrieval system.The humming music data used in the experiment were recorded in the standard studio by the phonetic Technology Laboratory of Suzhou University using ProTools audio recording equipment.10 experimenters of different genders each hummed 20 segments,averaging 8 seconds each,and the music library contained 800 pop music downloaded online on an 8s length.The experimental results show that compared with the existing traditional humming music retrieval system,the retrieval accuracy is improved by 5%,and the retrieval time is reduced by 9.5%.