The challenge was divided into six parts, each of them consisting of a different technology. There was a cash prize of $10000 for the breakage of each technology provided you agreed to be bound by a Non Disclosure Agreement.
Four of the challenges (named A,B,C and F) were watermarking technologies. The two others were probably essentially digital signature technologies.
We were primarily interested in the watermarking technologies. The information given for the other two was much less than what could have been expected in a fair challenge.
Each watermarking challenge included three songs, two copies (one marked and one unmarked) of a first song, and a marked version of a second. The goal of the challenge was to remove the mark from this second song. The success of the challenge was assessed by an ``Oracle'' which could receive songs through a web interface and indicate whether the attack was successful. The inner working of the Oracle were not well specified. From our experiments, it seemed that it was both checking for the presence of the mark, and also testing the ``quality'' of the music in some automated way.
The material provided is naturally largely below the usual settings of a cryptographic analysis. One could have expected to have a larger sample of songs to ease statistical analysis, a faster access to the Oracle or even the details of the marking algorithms.
In spite of this, we were able to almost fully analyze one of the schemes, the watermarking technology F, which we present and attack below.