Framework for Audio Fingerprinting based on Discrete Wavelet Entropy

Abstract

At the core of the presented system is a highly robust fingerprint extraction method which enables searching a large fingerprint database with only limited computing resources. Requirements for such systems include robustness to a wide range of signal distortions and availability of fast search methods, even for large fingerprint databases. In this paper an audio fingerprinting system is presented for song identification. For the high dimensional audio fingerprint data, audio fingerprint searching algorithm were proposed: an audio fingerprinting method based on DWE (Discrete wavelet entropy) with timbral features (MFCC and FFT) and an efficient indexing method for Audio fingerprint database using the filtering approach, known also as vector approximation approach which supports the nearest neighbor search efficiently. Spectral subband entropy is selected due to its resilience against equalization, compression, and noise addition. Region Approximation Blocks divides a highdimensional feature vector space into compact and disjoined regions. Each region will be approximated by two bit-strings according to the RA-Blocks technique.