I just had a sick thought.
I'm presenting tomorrow in the "VoIP Spam : Challenges and Solutions" panel, which means (of course) I'm writing it now. The other distinguished members of the panel appear are vendors, so my supposition is that they are selling solutions. Since I'm not a vendor, I'm writing about challenges. So, here I am thinking about VoIP Spam challenges, and I had a sick thought.
I remember hearing once that, after years of building SPAM filters for e-mail, a very effective method of determining SPAM was simply looking at people reporting SPAM from their inbox. If a bunch of people reported an e-mail as SPAM.... it was probably SPAM. Of course, this makes some sense.
Forget that approach for a second, and you are only using SPAM filters. Imagine that, to increase the chances of getting through the filter, you just changed spelling and smaller parts of the text. That would help get the SPAM through, but in general, there are only so many permutations because you are dealing with a fairly small field of possibilities, and in the end, you still had to have some sort of HTML link in there, and how much could you vary that?
My sick thought was that, with voice, I can vary it nearly infinitely such that it would be impossible from simple correlations to determine if two voice messages were the same. As a human being, I would hear them as the same message. As a computer, forget it. And even if the computers got better at correlations, I would simply need to add more noise to the message, knowing that human ears are really good at picking out the voice from the noise. Any Aerosmith fan will tell you that. This is really hard stuff. That takes away a very powerful tool that we have with e-mail SPAM.
On the good side, I have a brand new application for Amazon Turks! You could pipe all your voice mails through a real person who would throw away that voice SPAM. Don't laugh - it might be what this all comes to.