Study On Sample Based Music Information Retrieval

With the popularization of computers and development of the Internet, digital music spread widely. How to find out the needed music quickly and efficiently in the massive library, is a significant subject.Sample-based music retrieval is exploratory researched in this paper, music information retrieval system and automatic duplicate-eliminating database are realized. Main task contains two parts as follows.Firstly, algorithm of projections of spectrograms is improved, spectrograms of samples are mapped to vectors and feature points are extracted. The accuracy of retrieval is increased by selecting projection matrix, changing the number of matrix, frame length and frame shift. Indices are established using double hash structures. Retrieval speed is improved by the combination of coarse retrieval with accurate retrieval. Through experiments, the accuracy and efficiency of this system are satisfactory.Secondly, study on automatic music duplicate-eliminating. Using peak points and key notes as features, automatic duplicate-eliminationfor WAV and MIDI filesin high speed are implemented respectively.In conclusion, the sample-based music retrieval has a bright future, and the system established in this paper is of practicality.