G.729 and the welfare state
I was going to fisk the Government’s depressing sudden love affair with the discredited nonsense of “lie detectors”, but I see the Ministry has already done it. Go and read; it’s an instant classic. And as a bonus, there’s a great comment from the Great Simpleton, who you occasionally find in comments here, about some effects of telecoms infrastructure on the welfare state.
It’s certainly all nonsense – the 3G voice codec, AMR Narrowband, includes a band-pass filter between 200Hz and 3.4KHz, as do G711 and G729, so the markers VSA relies on, which are to be found between 8 and 12Hz, will be undetectable on any current mobile or fixed phone. Even the AMR Wideband high-quality voice standard will pass nothing – the band-pass for that one is 50Hz-7KHz. Any sound that does turn up at the VSA, therefore, is an artefact of some kind – a stray cosmic ray, or the acoustic echo cancellation at the local exchange going out of kilter when it produces the synthetic network noise to reassure you the line isn’t dead. (You might be advised not to Skype the benefits office – they’re considerably wider band when they are comprehensible at all.)
To expand on my comment over there, though, someone already markets a voice-stress analyser application for Windows Mobile smartphones. It’s probably mostly witchcraft and social engineering, but it’s very likely easier to do the opposite; either filter out the frequency band that is meant to be the marker, which could maybe sound weird or be too obvious if you could hear it at all, inject noise into that channel, or create a synthetic signal. That would be the hardest of the three to implement, but it would provide some interesting affordances – you could choose to sound more untrustworthy. If you could hear it, that is.
The only thing this achieves, then, is to deny some people their bennies entirely at random. Which is, of course, a highly political act.
Update: See here
-
1
Pingback on Dec 9th, 2008 at 10:18 pm
[…] have a “science” tag?, class, electronics, hacker, politics, programming OK. So we looked into voice stress analysis and the world telecoms infrastructure. And we concluded that proper VSA – the sort with the peer-reviewed scientific papers an stuff – […]
-
2
Pingback on Dec 10th, 2008 at 10:51 pm
[…] so yer lie detector. It’s been something of a blogosphere hit. And in the comments, we have Nigel, who appears to know something about acoustic signal processing – in the sense of “makes […]
-
3
Pingback on Feb 7th, 2009 at 8:52 pm
[…] the business with Harrow Borough Council and the secret Israeli lie detector? We noted that it was unlikely that the signal it claimed to detect would be transmitted through the telephone system; an expert pointed out that it might be manifested in other ways; eventually we obtained a copy of […]
Leave a Reply
Recent Comments
World of Harrowell
BORISFEED- An error has occurred; the feed is probably down. Try again later.
Alternate Seat of TYR 
December 10, 2008 at 2:24 pm
A Yorkshire Ranter wrote: “It’s certainly all nonsense – the 3G voice codec, AMR Narrowband, includes a band-pass filter between 200Hz and 3.4KHz, as do G711 and G729, so the markers VSA relies on, which are to be found between 8 and 12Hz, will be undetectable on any current mobile or fixed phone.”
A don’t know whether he is right for any particular manufacturer equipment/software. However, in general, YR is talking through his hat on this particular aspect. But don’t worry too much: his overall conclusions are not necessarily wrong.
Anyway, technically there is no reason why lack of speech signal at 8..12Hz prevents the measurement of variations in tension (at 8..12Hz or whatever) in the muscules around the larynx.
The most likely effect of such changes in tension is variation in the instantaneous frequency of glottal excitation of the vocal tract (highly variable, though typically 50..200Hz: lower for males, higher for females and children). If YR’s argument on VSA were correct, by scientific equivalence, telephones would not work for men. It is, of course, well known that males can use telephones, even though the fundemental frequency of their vowel excitation is filtered out.
By measuring very accurately, the changes in pitch period from pitch pulse to pitch pulse (glottal epochs), the instantaneous pitch can be measured. Where that varies up and down (at 8..12Hz or otherwise), the frequency of variation can be measured on speech signals of telephone bandwidth, and even lower bandwidth.
The weakness in VSA, at least as I understand it, comes from two aspects: (i) the strict causation between lying and stress is not present, and many other factors affect stress; (ii) any causation between stress and the VSA measurement is highly variable, and dependent on (at least) the person, their mood, health and recent activity; use of normalisation questions (such as “are you John Smith?”) support only very limited correction.
Thus, although there (probably) is some causation (at least for a significant proportion of the population) between the VSA signal and lying, it is not at all accurate. Professional liars (in the sense of doing it for money, such as social security fraudsters) can often get away with it; the innocent who dislike being grilled, or dislike begging for handouts, are also likely to be under stress.
Finally concerning human speech, in my view, the human ear and human brain are likely, when properly trained, to be superior to any machine in detecting lying; this is especially so when in conversation, with the possibility of supplementary questions. The use of a machine is, again in my view, likely to be a get-out so that social security staff (in this case) can wibble “Computer says no.”
Best regards
December 10, 2008 at 2:59 pm
Well, if your AMR Narrowband implementation isn’t filtering everything under 200Hz, it’s something else, not AMR.
I understand you to mean that the markers used in VSA could be measured indirectly – is that correct?
Meanwhile, I actually got the patent for the tech in question and a) it doesn’t work that way and b) I doubt it works at all.
December 10, 2008 at 3:47 pm
YR writes: “I understand you to mean that the markers used in VSA could be measured indirectly – is that correct?”
‘Indirectly’ is not the word I would use. It’s just that what is measured is a more complicated than an 8–12Hz signal in the speech. It’s more like frequency modulation (FM) than amplitude modulation (AM) or single sideband (SSB).
YR also writes: “Meanwhile, I actually got the patent for the tech in question and a) it doesn’t work that way and b) I doubt it works at all.”
In parallel with YR’s commend here, I have also argued against both of these over on his main website. I just don’t believe it works well enough to be operationally much use.
Best regards
December 10, 2008 at 4:00 pm
Meanwhile, it seems that the makers are sockpuppeting this blog and also spamming Wikipedia. More soon.