Audio Fingerprinting MUMT 611 Ichiro Fujinaga McGill University
MUMT611 Fujinaga 2 / 11 Introduction Fingerprints uniquely identify people Audio fingerprints aims to uniquely identify a piece of music from a short excerpt of the music Other names: Acoustic fingerprinting Content-based audio identification
MUMT611 Fujinaga 3 / 11 Applications “The popular social networking site MySpace.com announced Monday that it has licensed technology [Gracenote] that will help it prevent unauthorized copyrighted music from being posted to MySpace users’ pages.” Macworld (2006/10/06) “Adding missing album art: With the increased emphasis on album art, Windows Media Player 11 also ensures that missing album art isn't a problem. Most album art can automatically be populated in the background using the advanced audio fingerprinting capabilities in Windows Media Player 11.”
MUMT611 Fujinaga 4 / 11
MUMT611 Fujinaga 5 / 11 Commercial products Gracenote Gracenote M2any M2any Audible Magic (Muscle Fish) Audible Magic
MUMT611 Fujinaga 6 / 11 Basic framework (Cano et al. 2005)
MUMT611 Fujinaga 7 / 11 Challenges Variance Compression Distortion Noise Efficiency Encoding Loopkup Database size Search algorithm Music High dimensionality GOALS Robust Compact Fast
MUMT611 Fujinaga 8 / 11 Extraction Fingerprint extraction (Cano et al. 2005)
MUMT611 Fujinaga 9 / 11 Searching Euclidean / HMM sequence Pre-computed distances Multi-staged searching (coarse to fine) Indexing Candidate pruning Table lookup
MUMT611 Fujinaga 10 / 11 Table lookup database (Haitsma et al. 2002)
MUMT611 Fujinaga 11 / 11 References Cano, P., E. Batlle, T. Kalker, and J. Haitsma A review of audio fingerprinting. Journal of VLSI Signal Processing Systems 41 (3): 271–84. Haitsma, J., and T. Kalker A highly robust audio fingerprinting system. Proceedings of the International Conference on Music Information Retrieval. 107–15.