Reliable Automatic Recognition for Pitch-Shifted Audio

In this paper we present the difficulties involved in recognizing pitch-shifted audio, discussing its usage and implications. In the framework of audio fingerprinting techniques, we address the issue by proposing two complementary solutions, both of which can be exploited without affecting an existing reference database. As fingerprint algorithm, a well-known, robust method is employed in a modified version and providing insights on its peculiarities. Tests were carried out on a vast song library and exhibit an excellent success rate, up to considerable distortion magnitudes.