Panako: a scalable acoustic fingerprinting system handling time-scale and pitch modification

In this paper a scalable granular acoustic fingerprinting system robust against time and pitch scale modification is presented. The aim of acoustic fingerprinting is to identify identical, or recognize similar, audio fragments in a large set using condensed representations of audio signals, i.e. fingerprints. A robust fingerprinting system generates similar fingerprints for perceptually similar audio signals. The new system, presented here, handles a variety of distortions well. It is designed to be robust against pitch shifting, time stretching and tempo changes, while remaining scalable. After a query, the system returns the start time in the reference audio, and the amount of pitch shift and tempo change that has been applied. The design of the system that offers this unique combination of features is the main contribution of this research. The fingerprint itself consists of a combination of key points in a Constant-Q spectrogram. The system is evaluated on commodity hardware using a freely available reference database with fingerprints of over 30.000 songs. The results show that the system responds quickly and reliably on queries, while handling time and pitch scale modifications of up to ten percent