Music Fingerprinting Retrieval System Based On Perceptual Hashing

Recently, contented-based music information retrieval has became a research hotspot. Its mainadvantage is that it can find the music metadata from the own characteristic of the music signal,without any music text information. It has great applications. Contented-based music informationretrieval system can be divided into two parts: music feature extraction (audio fingerprintingextraction) and music retrieval. Perceptual hashing method is largely used in audio fingerprintingextraction. The experiment shows that it can get robust fingerprinting using this method.This thesis introduces the background knowledge and the state of the contented-based musicinformation retrieval. It mainly introduces the improved fingerprint extraction and search algorithmand analyses the experiment result. The main works are as follows:(1) Music information retrieval using perceptual hashing method is described. Perceptual hashused as fingerprinting is more robust. Perceptual hashes are extracted in consider of human auditionsystem. According human audition system, a series of threshold is calculated. The point that islower than the threshold is filtered.(2) In the process of retrieval, the priority of the sub-fingerprinting is set. The higher thepriority is, the higher the possibility that the sub-fingerprinting will be user as index is. In theprocess of audio fingerprinting extraction based on energy difference, the distribution of energydifference is Gaussian. A pair of threshold will be calculated. The number of the points outside thetwo thresholds will be the priority.(3) A series of experiment is done according to the improved audio fingerprint extraction andretrieval algorithm presented in this thesis. Those experiments show that those improved algorithmsare efficient.