I noticed a lot of questions here on the capacities of the Google/YouTube "Content ID" system used to recognize infringements. The following analysis is a bit old, and not particularly scientific, but it will give you a better idea of its recognition capacities:

http://www.csh.rit.edu/~parallax/