[messaging] Short Auth Strings

[Peter Gutmann]

Daniel Kahn Gillmor <dkg at fifthhorseman.net> writes:

>does "the uncanny valley" apply to audio?  what about audio over noisy/choppy
>channels, when users are used to "filling in the gaps" just to get on with
>their calls on a flakey network?

The human mind is incredibly good a reconstructing speech from almost any 
level of noise, from the well-known cocktail party phenomenon (the source 
discrimination problem) through to extremes like sinewave speech.  There's 
some examples of sinewave speech at 
http://www.lifesci.sussex.ac.uk/home/Chris_Darwin/SWS/, first play the SWS 
version (which sounds like Martian), then play the original, then play the SWS 
version again.

>But anyway, this seems like conjecture now.  Does anyone know of studies or
>work in this vein?

That's the problem with audio authentication mechanisms, people just assume
that they work.  I'm not aware of any studies in either direction, but from
what we know about human audio processing, it seems like something that's
inherently spoofable.  More specifically, I'd expect some sort of power-law
distribution for spoofability, with the long tail then being overwhelmed by
false positives.

Seems like something that could be pretty easily Turing-tested... hmm, now
there's an idea... if no-one's aware of any studies then please stand by...

Peter.