We're using the python port of audfprint though: https://github.com/dpwe/audfprint
And yes -- we have a big bucket of known ads, and a big bucket of archived programming, and we're comparing new content with the appropriate buckets.
NOTE: we are talking about ways to release the fingerprints of all the TV we archive so anybody can "search" TV for copies of sounds.
http://www.fec.gov/pages/brochures/notices.shtml#appearance
e.g.
"I'm Ted Cruz and I approve this message"
"This message paid for by the Americans for A Better Tomorrow Tomorrow"
There are very, very few non-political commercials that have this kind of phrasing.
edit: I'm only referring to how political commercials are different from non-political commercials. Obviously, the above heuristic has not much value in determining whether a particular ad is the same one as one already recorded and identified by the Archive.
They don't say how political ads sound different from other TV - I guess it works? Or maybe they are just using it to check if they have that ad already like how Shazam works?