Re: Voice Stress Analysis of Debates?
At 09:23 AM 10/6/96 -0700, Dale Thorn wrote:
jim bell wrote:
This reminds me... Years ago, somebody developed a technique called "Voice Stress Analysis," which was supposed to detect small variations in a person's voice in response to stress. Not exactly a lie-detector, but it was supposed to do nearly the same thing. Does anybody plan to analyze the debates for stress? Is there software to do this? (Tried to do a web-search; didn't see anything.)
Pardon me for butting in. Some remailer says I post too much, and I should cut out some. They didn't say which posts I should cut out, tho'.
Cryptography is, basically, the area of hidden meanings. In a sense, a voice-stress analyzer is intended to seek out hidden meanings in a person's statements. So I'd say it's as on-topic as most of the stuff 'round here.
Anyway, as I understand it, the current technology in voice/stress analysis goes way beyond the polygraph at its best (current) level of technology. There was at least one agency that did voice analysis during the Simpson debacle, and the results were (as I recall) very promising. Apparently, a person who can beat a polygraph cannot beat a voice/stress analysis.
One thing I wonder is this: Can the stress indications be removed from a voice-containing signal by some sophisticated DSP processing? Just look for whatever effect that indicates stress, add it in equal and negative amounts to eliminate the apparent stress, etc. It might not make sense for anything less "critical" than debates, but if the control of the debates is as monopolized as we think it is, it is reasonable to think that debate participants would insist on a certain level of control over the audio signal.
BTW, the rule for these debates (as all presidential discourse) is: "He's lying." "How can you tell?" "His lips are moving."
Well, I sorta assumed this. But I would still be interested to get some kind of quantitative feedback on the debates. If anything, the ABSENSE of the news media's attempting to use voice-stress analysis is telling. Jim Bell jimbell@pacifier.com
jim bell wrote:
At 09:23 AM 10/6/96 -0700, Dale Thorn wrote:
jim bell wrote:
This reminds me... Years ago, somebody developed a technique called "Voice Stress Analysis," which was supposed to detect small variations in a person's voice in response to stress. Not exactly a lie-detector, but it was supposed to do nearly the same thing. Does anybody plan to analyze the debates for stress? Is there software to do this? (Tried to do a web-search; didn't see anything.)
One thing I wonder is this: Can the stress indications be removed from a voice-containing signal by some sophisticated DSP processing? Just look for whatever effect that indicates stress, add it in equal and negative amounts to eliminate the apparent stress, etc. It might not make sense for anything less "critical" than debates, but if the control of the debates is as monopolized as we think it is, it is reasonable to think that debate participants would insist on a certain level of control over the audio signal.
Well, it's been about 20 years since the Stockhausen/Soundstream digital reprocessing of the Caruso recordings (other artists as well), and it would make sense that "they" can do exactly what you suggest, and very well indeed. If you recorded the debates yourself, and if you could do the analysis (much easier than faking anything), you'd have something to compare to the talking-head versions from TV. Unfortunately, the major media will not likely present any voice-analysis info unless it's done by one or two of the nation's top labs, which (it goes without saying) get most of their work presumably from government agencies.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ SANDY SANDFORT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C'punks, On Sun, 6 Oct 1996, jim bell wrote:
One thing I wonder is this: Can the stress indications be removed from a voice-containing signal by some sophisticated DSP processing?...
Nope. The PSE, and voice stress analysers in general, measure stress by the ABSENCE of micro-temors in the voice. S a n d y ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Jim Bell <jimbell@pacifier.com> writes:
One thing I wonder is this: Can the stress indications be removed from a voice-containing signal by some sophisticated DSP processing?
Not very sophisticated, but cheap current technology: high end PC soundcard, dictation software, to convert speech to text, then speech synthesis software to put that back into speech. Of course pauses in speech, and the actual words and positions taken show through, and these doubtless contain cues, as does body language. I thought that perhaps speech synthesis (so that you could type) or both speech recognition, followed by speech synthesis, together with PGPfone, and a high bandwidth TCP packet `remailer' might be a fun application. Plus a gateway back in to the phone system, paid for by ecash (double blinded ecash). All doable currently, I think. With sufficient users of the remailer, you should be able to get packet lag down, and still converse relatively interactively anonymously. Better than a pay phone, your link could be encrypted to the remailer with a forward secret protocol (PGPfone is forward secret), even if the recipient was using an ordinary phone. Talking about attempting to discover whether politicians are lying (apart from `at all times by definition'), it seems that even if a highly accurate lie detector were developed, puppet politicians could be kept ignorant, and fed the info to discuss, so that they could tell untruths without knowing they were untruths, and so better pass analysis. Also some pathological liars apparently can do well at lie detectors because they are so used to lying that they can lie with no compunction. Successful politicians would be selected from those who possess such qualities. (One suspects most are pretty good already, in any case) Adam -- #!/bin/perl -sp0777i<X+d*lMLa^*lN%0]dsXx++lMlN/dsM0<j]dsj $/=unpack('H*',$_);$_=`echo 16dio\U$k"SK$/SM$n\EsN0p[lN*1 lK[d2%Sa2/d0$^Ixp"|dc`;s/\W//g;$_=pack('H*',/((..)*)$/)
participants (4)
-
Adam Back -
Dale Thorn -
jim bell -
Sandy Sandfort