Join 3,551 readers in helping fund MetaFilter (Hide)


Electrical fluctuations as a watermark for audio and video recordings
December 12, 2012 3:38 AM   Subscribe

Audio recordings usually include a low-level background noise caused by electrical equipment. The hum contains small frequency fluctuations which are propagated consistently over entire power grids. By storing the pattern of grid-wide fluctuations in a database forensics experts are able to use the hum as a watermark. This can determine when the recording was made, where it was made and whether it was recorded in a single edit.

This technique of Electrical Network Frequency analysis or ENF was first used by Catalin Grigoras about a decade ago - and as described in this Wired article. He uses the forensic tool DC Live Forensics to achieve the sort of analysis that has probably stretched your credulity in umpteen episodes of "CSI". More recently Ravi Garg has looked at the technique of analysing video so as to pick up the same watermark as transmitted by light bulb fluctuations.
posted by rongorongo (43 comments total) 34 users marked this as a favorite

 
This technique has already been utilized to determine that every single recording that pop chanteuse Madonna ever made contains a minimum of 400 punch-in edits on the vocal track.
posted by flapjax at midnite at 3:44 AM on December 12, 2012 [11 favorites]


"For the last seven years, at the Metropolitan Police forensic lab in south London, audio specialists have been continuously recording the sound of mains electricity." Apparently Warp have given them a five-album deal; they'll be going into the studio next year with Pan:Sonic's Mika Vainio producing.
posted by Prince Lazy I at 3:50 AM on December 12, 2012 [27 favorites]


Apparently Warp have given them a five-album deal; they'll be going into the studio next year with Pan:Sonic's Mika Vainio producing.

Ooh! New music! link

My cats (and wife) will not thank you for this.
posted by sebastienbailard at 3:58 AM on December 12, 2012 [2 favorites]


Interesting! Recording studios etc could easily overcome this problem by using battery powered equipment. Are they?
posted by Termite at 4:00 AM on December 12, 2012


I'd guess that if the audio engineer doesn't hear the artifacts, he or she's got no reason to try to make them go away.
posted by sebastienbailard at 4:05 AM on December 12, 2012 [1 favorite]


Recording studios etc could easily overcome this problem by using battery powered equipment.

That'd get a wee bit pricey, don'tcha think?
posted by flapjax at midnite at 4:16 AM on December 12, 2012


Like bullets, awesome.
posted by iamkimiam at 4:22 AM on December 12, 2012


Recording studios etc could easily overcome this problem by using battery powered equipment.

That'd get a wee bit pricey, don'tcha think?


No, why would it? You pay thousands, if not millions, of dollars on recording equipment, you aren't going to give it clean electrical as protection? I'm frankly surprised this stuff isn't protected from grid fluctuations already.
posted by DU at 4:23 AM on December 12, 2012


It's not so much grid fluctuations as (according to the article) the inaudible hum from nearby mains wiring being picked up by the microphone.

Presumably a filter that removes a narrow zone of frequencies around 50Hz would break this technique.
posted by pipeski at 4:26 AM on December 12, 2012


...you aren't going to give it clean electrical as protection?

Most professional studios do use power stabilization devices and really good, high-grade electrical cabling and hospital-grade outlets/multi-tap outlet boxes, etc. But, batteries to power all the gear in a studio would be crazy expensive, really bad for the environment and, um, not necessary unless your recording is gonna like, go to the police as evidence or something! Haha!
posted by flapjax at midnite at 4:31 AM on December 12, 2012 [5 favorites]


if you've had a widespread and long-lasting power outage you can hear the absence of this sound pretty profoundly.
posted by ennui.bz at 4:32 AM on December 12, 2012 [4 favorites]


So, then, to make a recording look authentic, you just need access to the database.

Gee, I can't imagine law enforcement ever abusing that privilege.
posted by Malor at 4:32 AM on December 12, 2012 [2 favorites]


I can't tell if this article is supposed to render us in thrall to the Awwwesummm Powerzzz of the Governmentz or whether it's a promotional effort spearheaded by the PR department of some research institute but the broad unqualified claims made here are just nonsense on the face of it. Apart from the fact that it may be practically infeasible to obtain useful signal of sufficient length and quality, it isn't at all obvious how this helps to pinpoint the location of the recording other than "in England, perhaps". Notice how Dr. Cooper is careful to note that "various forms of analysis" were used in the single court case that's held up as evidence of its efficacy.
posted by deo rei at 4:33 AM on December 12, 2012


You pay thousands, if not millions, of dollars on recording equipment, you aren't going to give it clean electrical as protection? I'm frankly surprised this stuff isn't protected from grid fluctuations already.

I know for a fact that it often is, working in narrated television. You couldn't use this tech to find out when a studio recording was made-- everything's behind expensive equipment that regulates current.
posted by Mayor Curley at 4:38 AM on December 12, 2012


But, batteries to power all the gear in a studio would be crazy expensive, really bad for the environment and, um, not necessary...

Cleaning the power input for power hungry device needs batteries or large capacitors. Otherwise how are you going to fill in the dips?
posted by DU at 4:42 AM on December 12, 2012


deo rei: The BBC article seems to have been written to support a radio programme that is due to be transmitted this evening UK time. In terms of technical feasibility my guess is it would be reasonably easy to verify if the technique works: two recordings made in separate locations, but at the same time, using mains-powered devices should have matching patterns of background noise at 50Hz. Assuming the technique works then anybody, not just the police, would also be free to compile their own database of fluctuations.
posted by rongorongo at 4:50 AM on December 12, 2012


Cleaning the power input for power hungry device needs batteries or large capacitors. Otherwise how are you going to fill in the dips?

DU, reading this might help you to gain a little more actual practical knowledge about what you think you're talking about.
posted by flapjax at midnite at 4:59 AM on December 12, 2012 [7 favorites]


Wonder if they pick up any EVP's with this thing.
posted by charred husk at 5:18 AM on December 12, 2012 [1 favorite]


If only they'd used Monster cables, their electricity would have been cleaner!
posted by blue_beetle at 5:23 AM on December 12, 2012 [9 favorites]


But can they get rid of it, using something similar to noise cancelling technology?
posted by sfts2 at 5:33 AM on December 12, 2012


If this is the Georgia Power Company 60-cycle-by-courtesy hum, we know it well. Standard Physics 101 lab experiment to calculate the acceleration of a falling body due to gravity: one 1" steel ball bearing, two bare high-voltage AC conductors (aka "wires"), vertical, parallel, a little farther apart that the width of the bearing. Paper tape, vertical, between the wires and very close to one of them. Drop the bearing down between the wires. A regularly repeating electrical discharge takes place between the parallel wires every time the alternating current alternates; and the flashover goes through the bearing, wherever it is, because that's the path of least resistance. Burns little hole in paper tape.

The little holes are supposed to be spaced a regularly increasing distance apart because the ball bearing is falling faster and faster, hence can cover more distance between each current alternation, and by measuring these distances you can calculate exactly what "faster and faster" means, which is to say you can pretty well(*) quantify the acceleration due to gravity.

"Supposed to be" is key. Because the Georgia Power Company 60-cycle-by-courtesy AC oscillation is so erratic, the measured results are all over the place. Lab instructors tell students "If you've never encountered GIGO in the physical science before, get ready to shake hands with it today. You'll be graded on your experimental setup, your measurements, and your math. Your actual result for G is just for laughs."

(*) pretty well = not in a vacuum

So they're saying this erratic oscillation is enough to serve as some sort of location fingerprint? Sounds... optimistic to me.
posted by jfuller at 5:48 AM on December 12, 2012 [11 favorites]


So they're saying this erratic oscillation is enough to serve as some sort of location fingerprint? Sounds... optimistic to me.


I caught a little of the scientist talking about this on the radio - it seems it is so erratic, it becomes as "unique as a fingerprint". Although I was surprised that the background hum could be consistent across disparate locations - I'm assuming the lab is recording one reference hum; I would have guessed that the complexities of the power grid, involvement of sub-stations etc would lead to a different hum in different locations. Sounds too easy to challenge to be fully relied upon in court.
posted by khites at 6:10 AM on December 12, 2012


The location fingerprint is *very* broad jfuller: all you can say that a recording was recorded in the vicinity of a particular electrical grid. Britain is on a single grid & the US is divided into some small number of grids, so you can narrow down a location to a general geographic area but that's about it.

khites: the whole grid is synchronised (bad, bad things happen if power stations feeding the same grid are out of sync), so you would expect oscillations to be exactly the same across any individual grid. modulo speed of light style limitations.

What it does let you do with much greater accuracy is two things: locate a recording in time and give a certain amount of confidence that the recording hasn't been tampered with, because any editing would show up in segments that couldn't be matched, or appeared to come from different time periods.

Of course, anyone with access to the fluctuation database and sufficient signal processing chops could probably add an appropriate signal to any recording you like, so you still have to trust the police not to commit fraud in pursuit of a conviction.
posted by pharm at 6:11 AM on December 12, 2012 [3 favorites]


A lot of records have guitar amp hum. It's easy to tell if it's 50Hz (England) or 60Hz (USA). Using batteries wouldn't solve hum problems unless there was no AC power at all in the studio. I found light dimmers were the worst for causing hum. Guitar pickups are notorious for picking up hum from anything (see humbucker).
posted by bhnyc at 6:19 AM on December 12, 2012 [1 favorite]


"But when supply is greater than demand, the generators will slow down and the frequency will go down," explains Dr Cooper.

Wait... isn't this backwards? Demand acts as resistance on the generator, so if supply is greater than demand, then the generator spins faster (not enough resistance holding it back). Now, the operator will notice this imbalance and regulate the system accordingly (change gears, adjust air/fuel ratio, turn a peaking-generator off, or some other measuare will be taken). If such measures are taken then, yes, the frequency will go down. But only because supply is now less than demand and the generator (or generators) are working harder and slowing down as they try to keep up.
posted by molecicco at 6:36 AM on December 12, 2012 [1 favorite]


I heard you can avoid this sort of thing if you record all your guitar solos while skydiving.
posted by indubitable at 6:38 AM on December 12, 2012 [2 favorites]


> I heard you can avoid this sort of thing if you record all your guitar solos while skydiving.

so that's what "air guitar" means. I was so wrong!
posted by jfuller at 6:46 AM on December 12, 2012 [1 favorite]


This was all figured out by a goth girl/woman working in the basement right?
posted by Eyebeams at 7:02 AM on December 12, 2012


Burns little hole in paper tape.

Sets paper tape on fire in my case. We turned the spark timer on too soon before dropping the object. Also, we were supposed to be using the 10Hz setting instead of 60.
posted by hwyengr at 7:59 AM on December 12, 2012


The opening sentence in the first article is:

"A rape victim has come forward to the police. She says she has confronted her attacker and has secretly recorded him admitting his guilt."

I would have thought that this kind of covert recording would be made on a mobile phone or other portable device, rather than a piece of recording equipment plugged into a power outlet.
posted by Kiwi at 8:04 AM on December 12, 2012


Recording studios etc could easily overcome this problem by using battery powered equipment.

Actually, I was wondering about that, and the article wasn't clear enough to answer my question. Theoretically it should also be possible to detect fluctuations in a physically nearby but disconnected power grid, unless you have a really good Faraday cage. But I don't have the faintest idea what the tolerances would be for that sort of thing.

mobile phone or other portable device

See above.

the whole grid is synchronised (bad, bad things happen if power stations feeding the same grid are out of sync), so you would expect oscillations to be exactly the same across any individual grid. modulo speed of light style limitations.

If you have at least two "control" recordings, theoretically you should be able to triangulate sources of fluctuation, such that the sample recording you're testing acts as a third point and pinpoints general location. Again, I haven't the faintest, tolerances, etc.

This is brilliant and I'm sure they'll continue to refine it.
posted by dhartung at 8:06 AM on December 12, 2012


Yes but wouldn't a mobile phone MP3 be a bit too compressed and generally lo-fi to pick up the ambient hum fluctuations?
posted by Kiwi at 8:15 AM on December 12, 2012


I would have thought that this kind of covert recording would be made on a mobile phone or other portable device, rather than a piece of recording equipment plugged into a power outlet.

Doesn't matter. 60Hz hum doesn't come from the power supply, usually, it comes in via the microphone and other parts of the audio signal path.

If you get your cellphone (or for that matter a camcorder or portable field recorder or any other piece of battery-operated gear) too close to a mains cable or fluorescent light, you'll get lots of hum. Although it's not as obvious, there's probably some small amount of it down in the noise floor on almost any recording.

Now I'm personally pretty skeptical that you'd be able to get anything off of a cellphone recording, but that's because the digital compression and filtering that's done to the input signal would roll off the noise, not because the cellphone didn't pick it up originally.
posted by Kadin2048 at 8:18 AM on December 12, 2012


pharm - bad, bad things happen if power stations feeding the same grid are out of sync

Interesting. What sort of bad, bad things? Would you end up with varying currents as the power stations drift in and out of sync? How is this controlled for? (I have no idea why, but suddenly I'm fascinated by this)
posted by metaBugs at 8:28 AM on December 12, 2012


metaBugs: As I understand things, it's almost impossible for a generator already attached to the grid to drift out of sync: the rest of the grid is much, much more powerful than any individual generator & will simply drag it into sync if it starts to drift.

Really bad things happen however, if you start a generator from scratch and try and hook it up to the grid when it isn't in sync. The grid will drag the generator into sync, putting an emormous torque on the generator and huge current flows through the grid connection.

I've read of generators being ripped out of their foundations when this happens, but I've never actually seen any pictorial evidence!
posted by pharm at 8:50 AM on December 12, 2012 [3 favorites]


I've got the same question as molecicco here. The article says increased demand makes the frequency go up? That's certainly the opposite of what I would have guessed. Are there any EE's who can comment?

And I'm pretty sure Kadin2048 is right; the path is AC in wall > vibrations in air > microphone > recording. Batteries or other electrical isolation wouldn't affect it. You'd need to put acoustic protection all around the microphone except between it and the audio source.

But I really wish the article had given some idea of the scale: how loud is the sound, how much does the frequency vary, and over what time scale does that frequency variation happen? Heck, I don't even know what the units on that third item would be. Hz/sec? That looks weird.
posted by benito.strauss at 9:40 AM on December 12, 2012


"For the last seven years, at the Metropolitan Police forensic lab in south London, audio specialists have been continuously recording the sound of mains electricity." Apparently Warp have given them a five-album deal; they'll be going into the studio next year with Pan:Sonic's Mika Vainio producing.

Beaten to it.
posted by mykescipark at 9:51 AM on December 12, 2012


This sounds like criminal profiling, or forms of fire forensics, or all kinds of other things that may have a grain of truth but are nowhere as reliable as the people selling make them out to be. I’m sure people will go to jail based on this, and many years later be let out with an apology.
posted by bongo_x at 10:05 AM on December 12, 2012 [1 favorite]


Yes, increased demand makes the frequency drop. When the British grid's two largest suppliers went out within 5 minutes of each other in 2008, there was a frequency excursion down to 49.2Hz, which caused the grid to shed some load by cutting out power to a hew hundred thousand. It's described on Wikipedia.
posted by ambrosen at 10:10 AM on December 12, 2012 [1 favorite]


While the article is about the 50Hz hum in the UK (which I actually noticed when I was over there, and it's so strange how infra-sound or near infra-sound can affect the feel of any location), the 60Hz hum in the US is one of those strange things that you never really notice until it's not there.

As for "fooling" this technique, it's quite simple. Take a parametric EQ, and notch out the 49Hz to 51Hz frequency space. I do this anyway on my drum tracks, to give the kicks a better attack response (or something silly like that. I've had hours long arguments with people about the use of notched audio filtering, and I still come down on the side of intentional frequency notching).

Also, I am curious as to just how good cell phone recording microphones actually are. I did not think they had the frequency response low enough to actually pick up the 50Hz or 60Hz hum. Though I imagine they might be able to pick up overtones, should they exist, but I would think those would be lost in the noise floor or over powered by any other actual sounds.

Though, theoretically, I guess you are really listening for the silences more than anything else. Also, this would only work on single track recordings.
posted by daq at 11:10 AM on December 12, 2012


OK, let me get this right: they can tell where a recording is made by the electrical hum noise, but they still can't decisively tell which parts of a photo are manipulated ("photoshopped")?

Wow.
posted by IAmBroom at 12:47 PM on December 12, 2012


This sounds like criminal profiling, or forms of fire forensics, or all kinds of other things that may have a grain of truth but are nowhere as reliable as the people selling make them out to be.

e.g.?

In this case it's electrical engineers doing it. And there can be a stage where they superimpose signals on an oscilloscope. So it sounds like an objective, reality-based process, as opposed to "how does this evidence make me feel?"
posted by sebastienbailard at 2:40 PM on December 12, 2012


Here, for those who can get access to BBC Iplayer, is a link to the radio programme mentioned. The interview with Alan Cooper comes about 20 minutes in. He talks about FNF analysis as being used with recordings that are made with mobile phones - as well as those done with mains connected devices. His database of background sounds is apparently split into chunks of 1.5 seconds. It is not clear to me whether each of these 1.5 second chunks comprises its own fingerprint - or whether it just represents a single figure denoting the mean frequency during this period (I suspect the latter). He says "When the software searches the database it kind of reports in a probabilistic sense the chances of the pattern occurring by chance; usually it is many millions to one".

We don't get to know how long a recording would need to be in order to establish such odds. What the programme does illustrate is that it seems to be possible to time-stamp and integrity check a recording which much more precision than is the case when trying to calculate the odds that the person recorded is a particular suspect.
posted by rongorongo at 1:32 AM on December 13, 2012


« Older Sumit Suman recently visited a site, did not sign ...  |  Press X not to Die - one woman... Newer »


This thread has been archived and is closed to new comments