Research On Extracting Technique Of Audio Fingerprinting Based On Content

As the rapid development of computer network technology and multimediatechnology, images, video, audio and other multimedia information are moreconvenient to spread, especially in the mature of digital audio compression technologyand the emergence of large-capacity memory. With audio information on the Internetare too large, songs’ selection with hands has become impossible, which can bedirectly to promote digital audio fingerprinting technology of automatic musicidentification. Audio fingerprinting technology has attracted much attention as aseparate element in the audio format and without the additional data or informationembedding audio.In this paper, aim to content-based audio fingerprint extraction, based on audioproperties, from the three aspects of the main features of time domain, frequencydomain, time-frequency domain of algorithms for audio fingerprinting to study. Afterresearch and analysis of audio fingerprint algorithm on domestic and international, inthe frequency domain, due to the defects of the MFCC algorithm for high-frequencycharacteristics are not sensitive and the lack of dynamics, this paper proposed animproved MFCC algorithm. In addition, in time and frequency domain, fromimproving the robustness against the linear speed of the attack and reduction of thefingerprint storage space considerations, a fingerprint algorithm based on db4waveletstatistical characteristics is proposed.Through the bit error rate and robustness analysis of the MFCC improvedalgorithm proposed, the results show that the improved algorithm performance in avariety of attacks treated better than the traditional MFCC algorithm. In addition, theproposed audio fingerprinting algorithm based on db4wavelet statistical characteristics,not only the linear velocity attack has a good robustness, but also fingerprint storagespace has markedly less than the frequency domain audio fingerprinting.