I've been trying to think of how there could possibly be enough variation to fingerprint someone based on the hum caused by that 60Hz frequency noise. I've been in transmission control centers where they monitor, regulate and occasionally wet themselves over frequency shifts, and I've seen that the amount of variation needed to cause sheer panic is shockingly low..and it rarely ever happens for even a second. And those tolerances have been the same everywhere I've gone.
The frequency is synchronized across the whole grid.
The phase shifts, due to several factors (which way the power is going on the lines (treated as signal transmission lines), power factors of loads switching on and off, etc.) Much of this shiftig is local (motors on your transformer starting and stopping, etc.). Some of it is regional (for starters: the average across a distribution block of all those motor loads switching).
The combining of the varioius contributibutions to the phase offset is essentially linear. So if you have a recording system that is including power line hum and sufficiently stable on a tens-of-seconds time scale, the phase can be extracted and correlated with a recording from a nearby part of the grid. The closer they are (in electrical term), the stronger the correlation.
I could imagine the NSA recording this phase signal from one or several places in each city or rural region and archiving it, then using a cross-correlation against such a signal extracted from a recording. The amount of data to be stored and processed would be pretty small and a hit would stand out like a beacon.
First run against a national average (or several regional signals) to get enough of a hit to identiy the time of the recording. Then run against that time segment of the whole database of local samples to get a rough location. (With enough samples this should get you down to a "which cell tower" level.) Then see what suitable recording studios are in the identified region and look for other clues.
- Notch-filter out the power line frequency and its first few harmonics.
- Bandpass filter out the low part of the audio.
- Add in a small amount of hum of your own, with a pseudo-random phase jitter (and still more phase jitter on the harmonics). Be sure to use a set of pseudo-random generator that they won't be able to identify and cancel out - like by using several of them to continuously adjust the amount of phase noise added and such.
- Jitter the sampling rate.
- Re-record it with deliberate injection of a larger amount of real power line hum at a different time and location, before releasing the recording. B-)
Identifying edits in a recording consists of lookinf for a gross jump in the phase of the hum. Identifying the recording location from the pattern of small phase shifts (and other artifacts) in the power line signal is a much signal to find in a much larger amount of noise. I'm not convinced yet how doable it is. But with the above description of what I think they're doing, I expect a bunch of slashdotters will soon be playing with their audio cards, hacking up code to analyze recordings. B-)