A Framework for Robust Audio Fingerprinting

Abstract

We present a framework for audio fingerprinting, rather general in its essence, but especially tuned for being used in the context of broadcast monitoring. We efficiently implemented a robust fingerprinting algorithm and a suitable retrieval method. Ample sections are devoted to strategies for improving both the reliability and the speed of the overall system. The outcomes of plentiful experiments on a database of 100,000 songs are analyzed, and two common kinds of distortion (pitching and thermal noise) are investigated. To better drive design decisions, we also provide in-depth discussion on the scalability of the indexing algorithm.