Audio fingerprinting

Abstract

The aim of this dissertation is the study of the technology of audio fingerprinting in detail. This is a brief and to the point digital summary that is generated from an audio signal and is utilised to locate similar specimen that are stored in a database; and to, thus, identify an audio sample (Cano et al., 2005). Existing audio fingerprinting algorithms are reviewed along with their requirements and uses. This dissertation is also accompanied by a working audio recognition system that makes use of audio fingerprinting technology to identify samples of audio. Similar systems include the Shazam application from Shazam Entertainment, Ltd., Philip's Robust Audio Fingerprinting System, and Microsoft's Robust Audio Recognition Engine (RARE). The system accompanying this dissertation introduces an audio fingerprinting and audio recognition system that makes use of two types of time series processing, Fast Fourier Transform (FFT) and Symbolic Aggregate Approximation (SAX), to encode and fingerprint audio. The system also searches for unknown audio samples inputted by the user against its database of audio fingerprints. Furthermore, the application of each time series is compared to the other with respect to robustness, accuracy, efficiency, and speed. Through the various tests carried out, the system was shown to work effectively through the application of both forms of time series processing. But, it was noted that the Fast Fourier Transform (FFT) time series performs more efficiently and accurately than the Symbolic Aggregate Approximation (SAX) time series.