5 ms·

Any thoughts on how "unhackable" this system is?

dkokelley 14y ago
If the hum is inaudible to humans, but can be detected in the recordings, all that is necessary to create a plausible fake recording is to use a device that is incapable of capturing the hum, or severely compresses the hum so much that the hum is not distinguishable. This may not provide a means of proving the recording is authentic, but it prevents the defense from demonstrating that it was fabricated.
- levlandau 14y ago
  Hacking the system requires reproducing an identical "hum" and not necessarily compressing it. It seems difficult but I wonder if it is impossible for someone with a synthesizer and access to their database.
  - p0ckets 14y ago
    You don't need their database, you just need to record some background noise (assuming that you can replace the hum in your recording with another one).
    - levlandau 14y ago
      interested in the false positive (hum is present) in which case the hum you need to "replace" something with needs to have signal correlations that match the ones in the database they've built up.
madoni 14y ago
I imagine you could filter out the 48-52Hz range (probably less) and make your fake recording unverifiable. Taking the audio from the time you want to claim the recording took place and adding it shouldn't be too difficult either, though that might leave other clues that it's not genuine.
- levlandau 14y ago
  Right. I am more interested in the scenario where you fake something that is judged a genuine recording but is actually not.
- dkokelley 14y ago
  Another approach, dub the inverse wave of the real hum over your recording, which should (in my limited understanding) cancel the hum out, leaving a humless recording for a fraudulent hum to be overlayed.
- haldean 14y ago
  Except you would get the harmonics at 120Hz, 180Hz, 240Hz, 300Hz, etc.; there are some harmonics that are going to be within voice frequency that you wouldn't be able to filter out easily without taking out some of the body of the voice. It would (at the least) be obvious that the signal had been tampered with if it were missing all of the 60Hz harmonics.
- sneak 14y ago
  In the US, the hum lives at 60Hz. And of course as others have pointed out, there are harmonics.
glimcat 14y ago
The general premise is that there is an arbitrary, recoverable signal which has been convolved with the data signal. The generating function for this signal is a matter of record if you have access to the utility logs, but difficult to obtain otherwise without being physically at the time and place where the effect occurred. I strongly suspect that it can be compromised under both of the following conditions: 1. You take recording A, then you take recording B at a time and place which you want to assert that recording A took place at. You recover the thumbprint from recording B, suppress the thumbprint in recording A, then apply the thumbprint from B to A. This is not a trivial process, but you only really need a plausibly consistent result. A reasonably basic understanding of signal processing theory, a copy of MATLAB, and many pots of coffee should do the job. Then, you could automate most of it for the next guy. 2. You take many recordings at a series of locations of interest, while taking data about the power grid from nearby locations and from distribution nodes. You then attempt to predict the signal at a location from the characteristics of the surrounding area. This is almost certainly possible, as generalization from distribution logs to the local effect is what makes their fingerprinting technique possible in the first place. It is not a trivial undertaking, and it's questionable how well it would be generalizable. But at the same time, it's largely a question of if you want the data badly enough to do the legwork, and whether you have a reasonably functional understanding of machine learning.
- levlandau 14y ago
  Yes. This is what i had in mind. I agree. (In principle) the system can be compromised with some effort and skill.
- hatcravat 14y ago
  Regarding point 2, the article gave me the impression that they used historical measurements of the the mains frequency as part of the analysis. Since Britain is on a single grid, the local conditions shouldn't affect the recorded signal. I'm sure it is possible to find patterns (frequency drops slightly in the morning as the public utility tries to keep up with increasing demand), but you wouldn't know, for example that the utility overestimated demand on the particular morning that the audio was alleged to have been recorded (and thus that the frequency was actually higher). As for point 1, there are a number of plausible angles to approach this, but I think that the forensic adversary has a huge advantage: Synchronous detection. The approximate time of the recording is known, as is the historical record of mains frequency. That allows for the possibility of huge processing gain, which might allow for recovery even after the amplitude of the mains hum is filtered to below the quanta of the audio system. I almost think you might have to Fourier Transform the whole audio record and zero out any component at f_mains +/- delta (and harmonics). That, of course, would look pretty suspicious to a forensic analyst. Even that might not be enough if mains hum has a determinable effect on the data compression algorithm used to store the audio data. Edit: The point I was trying to make in the first paragraph is that if you wanted to forge a recording, you'd need to have the grid frequency data. Having it for anywhere would be good enough and not having it for anywhere simply wouldn't.

If the hum is inaudible to humans, but can be detected in the recordings, all that is necessary to create a plausible fake recording is to use a device that is incapable of capturing the hum, or severely compresses the hum so much that the hum is not distinguishable. This may not provide a means of proving the recording is authentic, but it prevents the defense from demonstrating that it was fabricated.

levlandau 14y ago
Hacking the system requires reproducing an identical "hum" and not necessarily compressing it. It seems difficult but I wonder if it is impossible for someone with a synthesizer and access to their database.
- p0ckets 14y ago
  You don't need their database, you just need to record some background noise (assuming that you can replace the hum in your recording with another one).
  - levlandau 14y ago
    interested in the false positive (hum is present) in which case the hum you need to "replace" something with needs to have signal correlations that match the ones in the database they've built up.

madoni 14y ago

I imagine you could filter out the 48-52Hz range (probably less) and make your fake recording unverifiable. Taking the audio from the time you want to claim the recording took place and adding it shouldn't be too difficult either, though that might leave other clues that it's not genuine.

levlandau 14y ago
Right. I am more interested in the scenario where you fake something that is judged a genuine recording but is actually not.
dkokelley 14y ago
Another approach, dub the inverse wave of the real hum over your recording, which should (in my limited understanding) cancel the hum out, leaving a humless recording for a fraudulent hum to be overlayed.
haldean 14y ago
Except you would get the harmonics at 120Hz, 180Hz, 240Hz, 300Hz, etc.; there are some harmonics that are going to be within voice frequency that you wouldn't be able to filter out easily without taking out some of the body of the voice. It would (at the least) be obvious that the signal had been tampered with if it were missing all of the 60Hz harmonics.
sneak 14y ago
In the US, the hum lives at 60Hz. And of course as others have pointed out, there are harmonics.

glimcat 14y ago

The general premise is that there is an arbitrary, recoverable signal which has been convolved with the data signal. The generating function for this signal is a matter of record if you have access to the utility logs, but difficult to obtain otherwise without being physically at the time and place where the effect occurred. I strongly suspect that it can be compromised under both of the following conditions: 1. You take recording A, then you take recording B at a time and place which you want to assert that recording A took place at. You recover the thumbprint from recording B, suppress the thumbprint in recording A, then apply the thumbprint from B to A. This is not a trivial process, but you only really need a plausibly consistent result. A reasonably basic understanding of signal processing theory, a copy of MATLAB, and many pots of coffee should do the job. Then, you could automate most of it for the next guy. 2. You take many recordings at a series of locations of interest, while taking data about the power grid from nearby locations and from distribution nodes. You then attempt to predict the signal at a location from the characteristics of the surrounding area. This is almost certainly possible, as generalization from distribution logs to the local effect is what makes their fingerprinting technique possible in the first place. It is not a trivial undertaking, and it's questionable how well it would be generalizable. But at the same time, it's largely a question of if you want the data badly enough to do the legwork, and whether you have a reasonably functional understanding of machine learning.

levlandau 14y ago
Yes. This is what i had in mind. I agree. (In principle) the system can be compromised with some effort and skill.
hatcravat 14y ago
Regarding point 2, the article gave me the impression that they used historical measurements of the the mains frequency as part of the analysis. Since Britain is on a single grid, the local conditions shouldn't affect the recorded signal. I'm sure it is possible to find patterns (frequency drops slightly in the morning as the public utility tries to keep up with increasing demand), but you wouldn't know, for example that the utility overestimated demand on the particular morning that the audio was alleged to have been recorded (and thus that the frequency was actually higher). As for point 1, there are a number of plausible angles to approach this, but I think that the forensic adversary has a huge advantage: Synchronous detection. The approximate time of the recording is known, as is the historical record of mains frequency. That allows for the possibility of huge processing gain, which might allow for recovery even after the amplitude of the mains hum is filtered to below the quanta of the audio system. I almost think you might have to Fourier Transform the whole audio record and zero out any component at f_mains +/- delta (and harmonics). That, of course, would look pretty suspicious to a forensic analyst. Even that might not be enough if mains hum has a determinable effect on the data compression algorithm used to store the audio data. Edit: The point I was trying to make in the first paragraph is that if you wanted to forge a recording, you'd need to have the grid frequency data. Having it for anywhere would be good enough and not having it for anywhere simply wouldn't.