Met lab claims 'biggest breakthrough since Watergate' Police scientists have hailed a new technique that recently played a pivotal role in securing a murder conviction as the most significant development in audio forensics since Watergate. The capability, called "electrical network frequency analysis" (ENF), is now attracting interest from the FBI and is considered the exciting …
I know what you mean.
A few years back, I worked in a small SME, and the monitor I had (an old 19" CRT) had a high-pitched whine. Unfortunately, I was the only one with hearing sensitive enough to detect it, so all my co-workers didn't believe me. One day, a client came to see us, and promptly asked me how on earth I could put up with such a noisy monitor! That *finally* got me a nice shiny new screen...
I guess it's feasible in theory, but the public and legal profession should ask the same questions they would of any new forensic technique:
- Has the science behind it been peer-reviewed?
- Have extensive trials been conducted to determine the limitations and causes of error?
- What are the known limitations and contra-indications for this technique?
- What's the expected error rate?
If they can't answer these questions, it's not scientifically valid, and any lawyer should be able to have it excluded as evidence.
The legal system doesn't go much for peer review.
Forensic science has up to now mostly been about bluster. That is only starting to change.
And your 100% right, the public and legal professions should be pushing for change, pushing for peer review.
That never stopped the use of Low Copy Number DNA analysis in the UK, in some cases to 100% positively match people who absolutely could not have committed the crime.
See also fingerprints, or hell, our colonial cousins with their lie detectors- magical thinking at its finest.
"Jesis said it was you wot dun it.."
I'm calling BS on this one. With the amount of noise suppression and compression going on in digital recordings, most of the signal is thrown out to begin with. Especially the bits you can't hear. Throw battery operated recorders on top of that and it seems even more preposterous. And then they have on file "deviations once every one and a half seconds for the last five years," which I don't understand. They have some variable recorded every second? That's a pretty long wavelength compared to the 44100 cycles / sec on a CD. They'll have to explain it a little better before anyone accepts this.
This is not BS, a friend told me about this some years ago, it is used by the police in France (not in court though). If you give them the time and date of a recording they can check it against their recordings. Mains hum is quite powerful, 30 years ago it was used to be used for touch sensitive switches
Here's a fairly full pdf of the research, so it would seem that the ENF signal does remain substantially the same over large distances.
http://www.forensicav.ro/download/2006-05-23%20ENF%20AES%20Paris%20Grigoras_.pdf
(Note El Reg... you mis-spelled Dr GRIGORAS's name wrong by one letter in your article :) )
You take the edited audio recording and play it through speakers in an acoustically quite environment, while recording the speaker outputs with a microphone. In this process, you lose the ENF signals in the edited digital recording and replace it with the ENF signals at the time of the rerecord. The claim that ENF signals can be imposed on recordings made with battery powered devices with no electrical connections to the power grid is plausible under normal circumstances. What would be interesting to know is whether battery powered digital recording inside a good shielded room (e.g. USC Series 26 solid wall, with 110 db attenuation) has been tested using ENF.
Worse, you could take the ENF signal for any time in the past and superpose that onto a recording in order to give it a fake "provenance". As you said, play the original back through speakers from battery-operated kit, also record with battery-operated kit, and mix the fake ENF signal at a level that will drown out any faint interference that your (well-screened) recorder might be picking up.
All that analog hardware may not be needed.
From my understanding of this the ENF is encoded in the data stream along with the audio/video, meaning it can be manipulated or filtered using the right program, I can already filter out the background hiss from an audio track.
Its cool, but it will only raise the bar a bit. Dumb criminals will overlook this, slack jawed jurists will accept it as Gospel and infallible so a smart criminal could frame someone with only a little extra effort.
Hmm, well this doesn't sound completely bogus. But I'm very sceptical how much use it is.
First of all most digital recording I've seen has between a 10-14 bit D/A converter. I'm sceptical that with filtering any signal would be large enough to register at all or if it did it would be so quantized by the sampling into 1 or 2 bits that it would be difficult to get much useful information out of it. Secondly it's rare to see a system that doesn't compress to MP3 or similar these days. I would expect that artifacts from that process would so completely swamp the signal as to make it useless .
Finally if this came up in court my first question would be to ask how accurate the match to the recorded value was, and to ask if there were any other periods over the previous few months that were within 20% of being as good a match. I find it very hard to believe that there wouldn't be many. I imagine you could discredit any such evidence very easily
It's not for recovering the audio- it's for recovering a characteristic set temporally-spaced peaks to use as a "fingerprint" to match against a record- much the same way as OB contribution links can be monitored by a very low data rate backchannel containing only a "fingerprint" data feed, rather than the actual audio or video itself.
AC for obvious reasons.
Like so many forms of high-tech forensics, the techniques can be impressively sophisticated. But in the last analysis, the end result is someone standing in the witness box giving evidence. If I were on a jury, I'd be very concerned about how much reliance could be placed on salaried evidence. I have some technological and laboratory experience, and I know for a fact that most lab results are invariably far from the clear cut 'proof' that many people assume and popular TV shows suggest. Much of it is open to far wider interpretation.
The right to counsel for defence - free if necessary, is a long-established principle. Perhaps it's time for a similar right to a forensic defence too.
Whilst I would be quick to stand up for (sadly these days old British) freedom and liberty values to condemn anything I found that had the potential to be used and abused to undermine society in a very Orwellian ways, I am however totally ok with the idea society still very much needs to have policing. Sadly our world very much still needs it. So I would say I have to admit, I'm more than a bit in awe of the amazing work they have done here. It would never have occurred to me that mains interference could be used effectively almost like a background time stamp on recordings.
I'm guessing this methodology can also locate recordings to areas of the country, as well as to points in time. I guess the next extension of this work would be to increase the number of mains monitoring points around the country and then to also increase the sampling frequency to include other unique mains noise events. Higher frequency noise is less likely to travel far on the power lines, so that gives an even more exact location reference.
I think this is remarkable almost unthinkable research work. It is also a bit of a wake up call, as it highlights how new information can be extracted from old data. (This is something I've thought about in general with the way many people post their lives online, completely unaware of how a decade or more from now, new ways of processing what they say now may well be able to reveal a lot more about them).
I'm wondering with this technology they may even be able to eventually pickup mains noise events in mobile phone conversations and recordings, where the analogue parts of the phone encodes the local mains noise events, and then as it is transmitted as digital, the rest of the signal path is immune to interference. That way with sufficiently dense mains noise recording around the country (which is easy and cheap to do) they could in time build up a country wide cellar like method to find the approximate location of where even uploaded videos were recorded around the country.
Very interesting and thought provoking work, although also a bit disturbing as it is wide open to be abused in the future in an Orwellian way as well, to add yet another way to spy on people. So yet again society takes one more step towards Orwellian spying. The more I think about it, the more I think an Orwellian world is sadly getting unavoidable. For a start a lot of sheeple don't see the technical details of what is possible, so they won't see the dangers until its way too late.
Oh well maybe Orwellian spying could solve a lots of crimes (that is their sales pitch), but on the other hand (the part they don't want to admit) is that it could (and most likely will be) used to subjugate, lie to and distract a whole society and to silently punish anyone daring to speak out against it all (which is exactly what every Police State in history has done). I think our future world is both solving crimes and subjugation, because the problem is once more behaviours are listed as crimes, that biases our world combined with abuse of technology towards the Orwellian Fascism that Orwell feared and exactly what we have been getting ever more warning signs of in the news (almost daily) over the past few years. Even local councils show what micro management control freaks they are, let alone the really power hungry control freaks who get to the top of the country. So the world will continue this slide towards them extending their control over us.
Thanks for posting this interesting news The Reg. You have added one more thing to think about!
Where will all this end?! Its amazing how our world is changing and so fast these days. Wow.
Truly, I have been pondering some similar issues myself.
A few golems for the gears:
How easy will this be to fake a signature or "prove" something. Other posters have already raised this issue. I won't go into specifics, but it is possible to create a false recording using this methodology. I imagine the timing scheme the police use is not secret, and you could essential fabricate evidence that would prove you were making a recording in the woods at the time of the murder. This signature would "prove" that this murderer couldn't have been killing his lover, he was taping deer in the park that day (assuming Mains lines nearby of course).
Fingerprints, recordings, even photos and video are no longer reliable.
Worse even DNA evidence can easily be tampered with by someone with less than a college education in basic lab work.
To make it simple lets say there are 26 different markers, and each has a letter. The victim has a ID of ABCXYZ - you dont need to find ABCXYZ together, you can grab these markers from many different people, until you basically just need to have all the markers. The person you want to frame has ABCXYZ markers you just make sure that all those markers are on the murder weapon.
That's really the problem with almost any kind of evidence like this, it is very useful because it can help us prove guilt, but mostly for those who are not criminal minded. Professional criminals seem to be VERY aware of these developments and seem to be able to exploit them to the fullest extent, and rarely get caught be them. When a big brother, or even just a well intentioned regime gets control of such a tool it will almost certainly be abused.
I hope that juries and judges will realize this and start to put more emphasis on character witnesses, motive and the like. It's probably no mistake that the founders felt that to be convicted of something at least two people should be able to testify they saw you doing it.
Another interesting thought - using atomic level microscopic discovery to find similar remnant data that can tell us a history. One idea Hollywood grabbed was the concept of using a laser to "sense" the vibrations of voices caught while pottery was made to play back the ancient conversations of the people who where speaking while it was made.
Or how about running all our old recording of politicians and religious leaders through a voice-stress analyzer, and combine that with analysis of the video feed for facial expressions etc. I bet money on what the discover will be: Politicians Lie a lot!
If a tape recorder is driven by a synchronous motor then whatever the mains frequency the tape will allways carry an apparently fixed common mode hum. I think all household recorders are of this type.
And surprise, research showed that the variations in frequency are the same in Glasgow and London. Yes, it is a National Grid.
"Something else for Google Street View cars to log?"
They don't, but it's also publicly accessible information so they could if they wanted; just as you you can if you gain the smarts!
So far, however, the only agency I'm aware of that is logging this is a quasi-governmental one... so whom do you trust?
"I'm guessing this methodology can also locate recordings to areas of the country,"
Don't guess, ask. Especially if you're going to post an essay.
The whole of an AC electricity supply grid has to be in the exact same phase across the whole of the grid, otherwise you get horrible circulating currents which destabilise the whole grid. This is why you need something like a DC interconnect between independent grids (e.g. France and UK).
So, I guess you guessed wrong. Sorry. If you need more detail, copsewood at 14:53 has it.
And while I'm here, whilst the underlying theory of identifying *when* (not where) the recording was made may have some plausibility, it is entirely reliant on having a known accurate timebase in the recording. And since when has a $2 audio recording device (dictaphone replacement or whatever) had more accurate internal timing than the grid? And as has already been pointed out, if the recording device's timebase is mains derived rather than independent, the exercise also fails (50Hz +/- nothing, apparently, all the time).
I too call BS.
The 50hz signal on the mains grid does fluctuate (slightly) all the time a little + then a little -.
(it basically has to do with the draw (load) on the grid constantly changing.)
The timing on the recording device is NOT relevant at all, it does not need a internal clock at all.
It just needs to record in real-time and play back at the same speed (all of them do this)
The police are building a signature of a series of changes. Your $2 recorder is affected by the pulse from these power lines as it records.
So let say (again simplified for discussion) that you make a 5 minute recording.
During the 30 second interval you get:
50.01Hz - 49.99Hz - 50.003 - 50.013 - 50.072 49.983.
This creates a UNIQUE enough signature to isolate the specific time frame when the grid did its frequency dance that specific way. Barring tampering or a conspiracy this would tell you exactly when any given recording was made.
As far as localization goes, that also is possible, but I think terribly impractical. ANYONE who has used a baby monitor, knows the kinds of devices that can introduce some feedback and noise to the power lines. So yes there are very unique high frequency signals being run on power lines, that would identify a locality.
HOWEVER, I think these high frequency signals travel so short (i.e. neighbors vacuum cleaner does not affect my monitor) that in order to collect this data and make it usful you would have to be sampling the power rates EVERYWHERE, or at the very least block by block. Not practical or economical.
Thank god the US military has known of the dangers of signals on power lines for decades. At least at their more secure locations power goes into a generator/turbine that turns a gear on another turbine (transforms the electrical power to kinetic energy and the back again) quite brilliant and no sneaky comms on the power lines!
Do try to be sure to remember that just because something is unfamiliar or seems implausible to you, does not mean it is BS. You should have a more solid, "I know this is wrong for this reason" Just saying it doesn't smell right to you, doesn't provide any counter argument to the article.
Sound On Sound ran an article on forensic audio back in January, including ENF:
http://www.soundonsound.com/sos/jan10/articles/forensics.htm
The clock accuracy in even cheap digital recording devices is way more accurate than would be required to record mains frequency with the precision needed.
It all comes down to the question of how to sway jurors and judges who have no scientific training?
And given that the defence can't afford its own scientific study, its own solemn looking man in spectacles, the prosecution will win.
It is like the old "fibre analysis" or "paraffin test" or "partial finger prints" or "DNA off the lavatory floor".
:snark:
I'm afraid that I'm still not clear about what they're actually analysing., though - fluctuations in power? How does this relate to a murder trial? Pardon my denseness, friends, this sounds intriguing but, at this time, quite opaque.
Regardless, hooray for the forensics experts.
I know crystals are accurate but I question your assertion re their prevalence in cheap high volume kit, where every cent counts.
If I was building a <b>CHEAP</b> (low cost, low functionality) high volume voice recorder why would I use a crystal rather than a cheaper and less accurate ceramic resonator (probably still accurate enough to drive a voice recorder if not a clock) or even something that derives its timing from an extremely low cost RC circuit?
re "Sound on Sound" article: so what. It was written by a policeman, you expect us to trust it as independent and unbiased? He does at least make it clear that the ENF technique has value in establishing whether a recording has been tinkered with (I think I can believe that).
I've been a juror in an inner city; and I can definitely confirm that 'expert witnesses' are taken with a very large grain of salt. Unless we know what they're talking about, they tend to be thrown out in the jury room - after a fair bit of furious argumentation. I've found that people underestimate just how sceptical jurors are.
"Anybody care what the cost of recording and storing this information is, and whether it's really not just a great big invasion of privacy? Just wondering."
Well, sample once every 1.5 seconds. That's (86400/1.5) 56700 samples per day. At, say, 8 bytes (double precision float) for value and 4 bytes (seconds since epoch), that's 680,400 bytes per day, or 248MB per year. Nope, don't care. :)
As for privacy, it's pretty tough to argue that the power company monitoring its grid is an invasion of privacy. The application in forensics has some dubiousness though.
You need a fairly precise voltage meter, a fairly accurate timekeeping device, and a logger that records timestamps and voltages. Say, you record single precision floats with unix timestamps, that's eight bytes per measurement, times the number of measurements (say, once a second), plus some overhead. That setup times the number of places where you want to record, which might be half a dozen or less for the UK.
Then multiply the cost of the device by some factor to make up for government inefficiency. Let's say, ballpark, two thousand pounds per device, installed at convenient police stations, around the country, assuming they have mains power. Or perhaps utility offices. Or something like that. Then there's some small change for someone to drive around and visit each device at least once a year to see & certify if it still works properly, gather up the data and stuff it in an oversized database in the lab.
In short, peanuts, as these things go. If you want to record that stuff yourself, sure, go ahead. It'll cost you less, even.
Continuing, and in comment to various other comments, this sort of thing is _of course_ falsifiable, and so you can use it to frame someone else assuming the prosecutors aren't doing that already. But the thing isn't that it can be broken, but more that most criminals are stupid and the rest of the crime is more sort-of ``casual'' as in perpetrated on the spot out of desperation or other equally pressing circumstances. Most of the time investigators wade through the shambles of human endeavour gone wrong rather than the results of deliberate manipulation. Which is why when that happens and it's done professionally enough nobody is the wiser that manipulation happens. Which suits the perps, spies and such, just fine. Yes, you can be framed and no, if it's done well enough it's not going to be detected. But the likelyhood of that happening is for by far most people well below the likelyhood of dying at the hands of a suicide bomber, which in turn is lower than dying by traffic accident. So. It's not really a concern.
Of course, if the forensics take themselves seriously as scientists they must of course include the possibility of malicious input. But since it's well known in computing circles and still everyone forgets about it, I'm not so sure the forensics people manage to pay more than lip service. Right up until we discover that forensics are just as secretive for much the same reasons as bankers. But before that, there's simply not enough information to go on, so nobody will be raising enough hell to improve the situation.
Still, to me this is another tool and as such it isn't much of a privacy concern just like collecting dna samples and fingerprints at the scene of a crime isn't. Something has gone clearly wrong and it's good show if the hopefully competent and professionally impartial officials manage to find out what happened, who did it, why, and have them explain themselves in front of a professionally impartial judge. That's how it ought to work. Even the recording of power fluctuations isn't much of a problem like recording wood growth rings histories isn't a privacy invasion. It's useful data to have around much like historical weather data is useful to have around.
Contrast with preventively treating everyone as criminals by storing their fingerprints, dna, enough personalia to legally impersonate, and whatnot else ``just in case'', in what then turns out to be leaky databases, or just plain lose the data on the train for good measure. Juicy targets for identity thieves indeed. And then complain crime still hasn't vanished, and try and continue with battling thought crime. It's already happening, next to conditioning kids it's alright to be surveiled and biomeasured all the time. For school lunch, for checking out school library books, and so on.
As already has been noted, in a way this is quite an interesting development. I also recall DSP trickery to record monitor glow scattered from a wall and turn it back into a mostly readable picture. Since this audio thing depends on previously ignored ``noise'' and turns it back into meaning, well, what else is there? The CCD elements in cheap multi-megapixel cameras are notoriously noisy. Wonder what one can learn from the noise in digital pictures. And next to that, steganography may see a comeback of sorts.
I'm I the only one thinking this is old news? ENF technology has been around for years. There is even an open source application made the National Dutch Forensic Investigation lab (NFI). It allows you to record your own reference ENF pattern. (http://sourceforge.net/projects/nfienfcollector/)
Many sites around the world have been providing these patterns to "CSI" type of units.
So we are looking at SAMPLING frequencies.
You know, Nyquist & all that.
Cheap DIGITAL recorders sample at ??? KHz (that's a thousands multiplier)
the mains runs at 50 Hz
My, a cheap dsp chip could easily identify & record variations with a phase change in the range of half the sampling rate.
No black magic, no BS. Also realise that we can pull signals 10s of dB BELOW the noise floor out of signals quite easily.
Good out of the box thinking from some one to realise what could be done.
Simples
Power companies have buried a hidden signal which is picked up by recording media *everywhere* in a country.
And (I suspect) the pattern of *variations* over a time period is what is compared to give the timestamp.
I know need to go and get my tongue unstuck from inside my cheek.
"You need a fairly precise voltage meter, a fairly accurate timekeeping device, and a logger that records timestamps and voltages"
Give it up already.
You don't need (or want) a voltage meter, precise or otherwise.
What you do need is a precise frequency counter able to measure 50Hz plus or minus not much with reasonable resolution and reasonable precision, and a means of recording (and timestamping with reasonable accuracy) the results. The power generation companies monitor the mains frequency data as a matter of routine; it's highly likely they record it too, but an individual with a frequency counter can make the same observations and should get the same results anywhere on the UK grid.
You also need some confidence in the assumption that the device that made the recording has a timebase that is accurate and reliable. I question how reliable this assumption is. An earlier comment claimed that the Sound on Sound article showed it was a valid assumption. The Sound on Sound article (written by a police forensics expert) provides no evidence whatsover that it's a valid assumption.
Still, it's clear that the tool has usefulness for showing that a recording may have been tampered with, which in itself may occasionally be useful.
Oh dear Rob, be careful.
Obviously the mains frequency varies a bit. The data is public e.g. the National Grid measures it and puts the last hour's data online at http://www.nationalgrid.com/uk/Electricity/Data/Realtime/Frequency/Freq60.htm
"This creates a UNIQUE enough signature to isolate the specific time frame when the grid did its frequency dance that specific way. "
It *might* create a unique signature, assuming the patterns are always unique and that the recording itself is accurate enough and long enough for a worthwhile match to take place. Let's take the uniqueness and length as read.
"The timing on the recording device is NOT relevant at all, it does not need a internal clock at all.
It just needs to record in real-time and play back at the same speed (all of them do this)"
Highly questionable. Something accurately crystal controlled would probably be OK, something with a cheaper (and less reliable) clock would be less reliable.
Once upon a time the mains itself was used as a clock (eg for synchronous motors in tape recorders) and in that case this technique is useless because as the mains slows down, so does the recorder. Hence it always replays at what appears to be 50Hz exactly, even though it wasn't 50.0000 Hz at the time the recording was made.
And that's without going into areas like lossy compression already mentioned by others, digital compression techniques which are routinely used these days reduce the volume of data to be recorded by effectively doing a spectrum analysis and discarding signals which you can't really hear (like background mains hum). There is no mains hum to analyze in the playback, because it wasn't recorded.
"One idea Hollywood grabbed was the concept of using a laser to "sense" the vibrations of voices caught while pottery was made to play back the ancient conversations of the people who where speaking while it was made."
Are you joking, or were they?
"Do try to be sure to remember that just because something is unfamiliar or seems implausible to you, does not mean it is BS. "
Do try to be sure to remember that if you know a little and reveal just how little you know, your credibility may be at risk. Usually no one will comment though, they'll just laugh quietly. Those who have an independent clue in the subject area are often reasonably well equipped to detect self-promoting BS such as we have from the police here; there may be a grain of usefulness in the technique but it sure ain't universally applicable.
A. Coward, BSc, MIET
"Proponents are optimistic about its uses in counter-terrorism investigations, for example to establish when suspects made reconnaissance videos of their targets, or to uncover editing in propaganda videos."
That would be all those propoganda videos made inside the influence of the national grid then would it?
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
