Audio Fingerprinting based on Wavelet Spectral Entropy

Abstract

Audio-Fingerprints (AFPs) are essential characteristics of digital audio streams used to score the perceptual similarity between audio signals. Audio-fingerprinting systems extract features from the signal normally on a frame by frame basis. In this paper, a robust audio-fingerprint (AFP) approach is developed based on spectral entropy in wavelet domain. To extract the fingerprints of a song, Shannon’s entropy is determined from the spectral coefficients of each one of the first 24 critical bands according to the Bark scale. The performance of this AFP system is evaluated on a music database containing various genres. The robustness of the system is validated through degraded music signals in 4 different ways: white noise addition, lossy compression, lowpass filter and resampling.