Audio Identification And Authentication Based On Digital Fingerprinting

Digital audio fingerprint is a robust content-based compact signature that summarizes acoustic relevant characteristics of an audio recording, it is typically used for automatic music identification and audio authentication.This paper gives a vision on the concepts, properties and frameworks of fingerprint extraction algorithms etc., summarizes most state-of-art audio fingerprinting algorithms used for music identification and audio verification. Adopts wavelet packet best-tree decomposition as the basic math tool, and further calculates different statistics from selected wavelet packet coefficients as robust audio fingerprint, which forms the basis of our three designed algorithms for audio identification and audio authentication.Algorithm 1 is used for audio identification, taking the ratios between the energy of every frame and the mean energy calculated from wavelet packet coefficients as a sequence of robust audio signature. From experimental results we see that this type of fingerprint shows great ability to distinguish different songs and exhibits strong robustness between the distorted and the original fingerprint even after severe audio signal processing like MP3 lossy compression, noise addition, resampling, filtering etc.Algorithm 2 uses the above energy ratio and standard deviation ratio calculated in the same way as two different signatures for content-based audio soft authentication. Experiments show that these two features can both differentiate content-preserving operations like MP3 compression, noise addition, echo, resampling, filtering etc and malicious operations like tempering, replacing, random cut/paste etc, and can precisely denote tempered local regions.Algorithm 3 takes a sequence of+1 and -1 formed by the sign of the energy sum of all wavelet packet coefficients in every frame as the robust signature for quality-based audio soft authentication. To differentiate acceptable and unacceptable operations, we adopt PEAQ(Perceptual Evaluation of Audio Quality) as a tool to estimate whether a certain operation causes severe audio perceptual quality degradation. Generally, -1 is used as the critical point, below which is viewed as unacceptable. Experiments show that in the sense of PEAQ, only MP3 compression at all bit rates is quality preserved, other operations including echo, equalization, resampling, filtering, pitching shifting, time-scale modulation etc all make the PEAQ value below -1 and are viewed as non quality preserved, thus triggering the detector and making the authentication failed.