Forensic Hum

[DF96]

See BBC report on mains hum as a forensic tool for authenticating sound recordings.

 

[bigwill]

That's very clever! Even a high-pass filtered signal could work due to the harmonics.

 

[DF96]

Now we know they are doing it, it might be possible to add some hum from a hum recording made at the time the main recording is supposed to have been made. To do this the 'native' hum would have to be very low so it was well hidden by the added hum. One of the snags with any forensic method is that Mr. Plod and his lawyers often end up believing that it is 100% reliable, and this makes them lazy and careless about doing proper police work.

 

[Pano]

Very clever! I wonder if there is a database for the US power grids?

I'll try to catch the Radio 4 program this evening.

 

[bigwill]

One could also add a subtle time varying pitch / speed shift to swamp the known frequency variations, to wipe any embedded "timestamp".

 

[Pano]

But wouldn't that show up as tampering?

 

[bigwill]

But wouldn't that show up as tampering?

Well I mean, from the perspective of the criminal you might want an audio recording to be impervious to this technique altogether

 

[Pano]

Arrggggghh! Listening to the BBC 4 program about forensic voice work. Now all I get is:
"Due to rights restrictions this part of the program is unassailable. Due to rights restrictions this part of the program is unassailable. Due to rights restrictions this part of the program is unassailable. Due to rights restrictions this part of the program is unassailable."

WTF? Been listening all afternoon. Oh, there it's back again.
Anyway, I guessed the voice in the lineup. #4.

Show came back just in time for the mains hum story. Good stuff!

 

[DF96]

I thought it was just things like sports or music which are limited to certain markets. Some Radio 4 programmes get repeated on World Service, although the BBC is generally losing interest in that because the government has stopped funding it.

 

[Pano]

It was strange, as it was just one story in the whole block.
I wonder who they are recording the mains frequency. The radio story mentioned once every 1.5 seconds. Would be interesting to know how they do it, and what format it's stored in.

Also fun to listen to the mix and how and where the voices were placed. The Beeb does good work.

 

[DF96]

My guess is some sort of data logger. 1.5 secs is interesting, because how do you determine the frequency of 50Hz to the required accuracy in such a short time? They must use some form of rolling average or low pass filter. Possibly a frequency multiplier - maybe using digital techniques but then it might be subject to noise.

It is actually a non-trivial problem!

 

[Pano]

yeah, that's what I was thinkin' just how do you set about doing this right?

 

[benb]

I recall a story about the US grid(s) being somewhat "deregulated" - the long-term stability (as used in many electric clocks, including electronic digital ones, to keep time) would go away, causing mains-based clocks to show the wrong time:

Power-grid experiment could confuse clocks - Technology & science - Innovation | NBC News

I don't know if they did that, or for how long, but I have no doubt there are millions of digital clocks still based on the 60 Hz power frequency for their timebase, and they're probably STILL being sold in Walmarts and other retailers around the US.

One could also add a subtle time varying pitch / speed shift to swamp the known frequency variations, to wipe any embedded "timestamp".

Well I mean, from the perspective of the criminal you might want an audio recording to be impervious to this technique altogether

It's a lot simpler than this - use an old-fashioned analog tape recorder, especially a cheap cassette recorder. Even the better studio analog recorders have playback variation of a substantial fraction of a second over a few minutes.

What makes this technology (as described in the story) possible (and what the reporter completelly missed, because he most likely doesn't have a clue about this - likely the police and engineers do, but forgot to say, or they intentionally did not say) isn't just the variation in the power line frequency, but also that over the last two decades or so digital recording has become ubiquitous and has replaced analog recording. It is usually crystal controlled, and so time and frequency reproduction are substantially more accurate than the power line itself.

 

[DF96]

Let's assume you need to know frequency to about 0.01Hz - it won't change much beyond about 0.5Hz except under fault conditions. Naively that means you need 100 secs. A frequency doubler gets that down to 50s. My guess is some sort of digital filtering to remove noise, combined with digital pattern matching to determine the 'instantaneous' frequency. Then low pass filter. Finally sample every 1.5s and write the result to a log file.

Whatever method is used would have to be used for the recording too, but there the hum will be buried under the signal. Fortunately it is a narrow-band signal. The mains frequency can't change suddenly because of the inertia of all the generators.

Finally, you have to do correlation analysis to determine the actual time - a bit like tree-ring matching.

benb:
Yes, the recording needs to be crystal locked. Can you imagine lawyers in court arguing about jitter on the measurement?

 

[Pano]

It's a lot simpler than this - use an old-fashioned analog tape recorder, especially a cheap cassette recorder.

That was mentioned briefly, IIRC. That the ubiquitous digital recorders in cell phones and the like now make this possible.

(and what the reporter completelly missed, because he most likely doesn't have a clue about this -

"She", actually. And it was alluded to. The story wasn't in depth enough to go into that sort of detail. Interesting too, was that they said it is still impossible to get a certain match from voice analysis. Human voice just change too much.