Robust Landmark-Based Audio Fingerprinting

A landmark-based Shazam-like audio fingerprinting system.

Description

This landmark-based audio fingerprinting system is able to match short, noisy snippets to a reference database in near-constant time.

This is my implementation of the music audio matching algorithm developed by Avery Wang for the Shazam service. Shazam can identify apparently any commercial music track from a short snippet recorded via your cell phone in a noisy bar. I don't have the database to check if my version is quite that good, but it is able to rapidly match and locate a poor-quality excerpt from within a database of (at least) hundreds of tracks.

See http://labrosa.ee.columbia.edu/~dpwe/resources/matlab/fingerprint/ for the "published" output of the demo script.

Notes for running under Windows (from Rob Macrae) are at http://labrosa.ee.columbia.edu/matlab/fingerprint/windows-notes.txt .

• Required Products: Signal Processing Toolbox
• MATLAB release: MATLAB 7.8 (R2009a)
• Other requirements: The demo code relies on mp3read.m to load MP3s, and relies on being able to call "curl" from the command line. Thus, the demo will only run on Linux or MacOS. The fingerprinting itself should run on anything.

[Sebastien Paris]

Nice application

[malli]