Reply #8: It might be feasible, actually.

[gulliver]

A couple of billion dollars would buy a lot of blades and disk farms. A systematic analysis of voice traffic can begin with simple, readily available data items such as phone number, geographic location of the phone, caller id, telephone directory databases. That would get you a lot before you even need to penetrate the call itself. You would know who talked to who when and from where.

Then the voice recognition does not have to be perfect enough for dictation. It only needs to be able to index the call by key words and phrases. The call itself could theoretically be stored, but might only be stored if certain business criteria were met. Once you get a basic system in place, then it is just a matter of scalability. I think this is probably quite a bit easier than cracking high strength encryption, steganography, etc.

I agree that we very likely don't have the ability to "Google" terrorists yet, but I don't think we are far from it.