If the Department of Homeland Security can analyze single spoken "red flag" words in a call, then why the hell cant the carriers do some sort of checksum of the call data coming through their pipes? I'm not saying to "listen in" like a wiretap, just sample bits here and there to build a, "audio-spam" profile that could be matched against other calls since robocalls should have essentially identical voice pattern signatures. Now sure, just like spammers always keep trying to evolve to get around the filters, I'm sure determined robocallers would try tricks like pitch shifting the voice, altering silence durations between sentences, etc... but that seems like it cause them more work to alter the calls (as opposed to dial & press play) than it would to make new match patterns to filter the evasion tactics. Plus the more contorted the messages become, the less effective they would be to any receptive audience... again, think of it like spam - a small % of people might respond to an email offer to buy Viagra, but most of those same people are going to ignore an email offering V1.4g.r.A! because the contortions make the message so much less trustworthy...
Also, as far as dealing with the robocallers who are caught, there should be massive fines AND jailtime just like email spammers. Give a large portion of the fines to the people who got called (to compensate them and promote reporting of robocalls) and also give the telecoms who gathered the data to make a prosecution a slice, so they have a real financial incentive to stop as many robocalls as possible and recoup some of the money lost on their audio-spam profiling operations.
And if the robocaller happens to be outside our jurisdiction, call in a drone strike and label them an enemy combatant ;)