Audio fingerprinting based on multi-resolution analysis of Discrete Wavelet Transform

In this paper, two new fingerprinting techniques based on approximating octave frequency bands using multi-resolution analysis of Discrete Wavelet Transform (DWT) are presented. In first technique, energy difference at several frequency bands is used for deriving feature vector. In the second technique another feature vector including mean, Zero Crossing Rate (ZCR), normalized first moment and flatness of coefficients of each frequency band is computed. By modelling these vectors, two different fingerprint blocks are obtained. The robustness and discrimination power of proposed techniques are evaluated and compared to those of traditional PRH algorithm for audio fingerprinting.