Audio fingerprinting based on analyzing time-frequency localization of signals

Abstract

Due to the desired non-invasive property, fingerprinting is considered to be an alternative to achieve many applications previously accomplished with watermarking. Some techniques for audio identification or retrieval have been proposed in the literature. However, few of them were done by analyzing the time-frequency variations based on a transformation with efficient time-scale localization. In this paper, we shall investigate the characterization and recognition of audio based on time-frequency analysis of signals. One dimensional continuous wavelet transform will be adopted to capture the time-frequency variations of audio. Based on the multiresolution structure of an audio, two fingerprints are created for authentication and recognition purposes, respectively. Experimental results have demonstrated the performance of the propose method.