The Design And Implementation Of Music Retrieval System Based On Spectrogram Feature Extraction

With internet applications becoming increasingly more, the requirements of multimedia and the requirements of music search are more important.The traditional method of music search is based on text,and it is only according to the external text message about the music,such as the singer,the title and the songs introduction.Finding the results of this approach depends on the accuracy of the text matching algorithm,and it also depends on the accuracy of the text's description. Most of these descriptive text edited by hand, when the amount of music is huge,it is easy to make a mistake.The description text can not correctly represent the content of the music itself, which also affect the results of the query.In order to solve the problems about text-based way of matching, the paper proposes a new way of thinking. Uses a content-based music retrieval way, by extracting music files spectrogram,and then select the location as the music file information, and the combination of feature points to form a music file fingerprint information.According to this model and algorithm, the paper presents a musical fingerprint extraction algorithm, to ensure the uniqueness of music fingerprint.The paper also proposes a fingerprint through a music, and music library built music-based fingerprint algorithm and fast query methods, and it also analysis the query speed and accuracy through a test.Meanwhile, the paper alseestablishes a scalable fingerprint-based music retrieval system, and it is easy to extend the amount of music fingerprint. The music search system using the network communication, more than one query in parallel query server service, the results aggregated to sort out a dedicated server. The paper analysis of the bottlenecks in the system and proposes an effective solution.