vega.rd.no

Master's thesis: Combining Audio Fingerprints

Monday June 16, 2008

On June 3rd I delivered my master’s thesis. Hopefully, sometime in September I’ll receive my diploma with the title “Master of Technology/Civil engineer” (direct translation from Norwegian). In reality, my English title will be “Master of Science.” So now that I’ve spent a week eating cake, it is time to publish the thing.

Anyhow, since my thesis won’t be published by NTNU before it has been graded, I have to publish it here. So here you go, one master’s thesis, coming up. The associated code (minus the database, which was nearly 2GB) is also available. I’ll remove the my download links once NTNU has published it.

There is no front page or problem description, as that is added automatically when printing through NTNU.

In summary, the thesis uses three audio fingerprint systems, libFooID, fdmf and another master’s thesis from NTNU as a basis for creating a more accurate fingerprint. On the test collection, the combined fingerprint is approximately 60 percent more accurate than any of the individual systems.