Audio Fingerprint Extraction Algorithm Based On Hash Function

In recent years, with the continuous development of mul timedia technology and the popularity of the Internet, many kind of multimedia applications are everywhere in our dail y life. Especially music-aware application, it is often des irable to be able to identify unknown audio files, such as the song’s name, singer, composer and so on. As the networ k environment is inclusive and complexity, audio files have usually been compressed, or edited, make audio files disto rtion and incomplete. In addition, during the transmission of audio may also produce noise, how to extract fingerprint efficiently and robust in different environment and identi fy a large number of audio files has become a major difficu lty.Audio fingerprinting technology is a process of differe ntiating audio files by extracting characteristic of the au dio. Audio fingerprint is a compact sequence generating by hashing compression. It’s a uniquely feature that can tell audios apart. In this paper, the development of the hash fu nction and its application in the field of audio fingerprin ts were summarized, elaborated the development of audio hash function from the traditional hash function to the percep tion hash function. Perceptual hash function inherits the a dvantages of traditional hash function, and also robust in content-holding operation 、sensitive in content-tampering operation by simulating the human auditory characteristics.Mel domain is a simulative model of the human auditory sys tem.I also proposed a novel hash construction method based on Mel filter that include perceptual characteristics. The algorithm also include the dynamic coefficients and the ave rage energy value. The novel method improve fingerprint’s identification and robustness. The Haar wave transform has been used to reduce the volume of the audio fingerprint.For the proposed new fingerprint algorithm, this paper also proposes a novel two-factor matching algorithm, compar e the similarity of two audio fingerprint from two dimensio ns, the direction and distance, to improve the efficiency a nd accuracy in matching. Experiments were taken to test the new proposed algorithm. Experiments show that the new fing erprint has a higher precision with smaller volume, effecti vely reduce the space of fingerprint database. Two-factor c oefficients in matching process can improve the fingerprintmatching rate and applicable in degraded audio file and no ise background scene.