[Discuss] Fidelity voice-recognition security?

[richard.pieri at gmail.com (Richard Pieri)]

On 11/23/2017 7:36 AM, Robert Krawitz wrote:
> Bit of a difference there, you agree?

Yup. Although the "we have no idea" bit is hyperbole because the reality
is we do know how unique fingerprints are. Francis Galton did
substantial scientific research on fingerprints back in the late 1800s.
He estimated the chance of two people having the same fingerprints at
around 1 in 64 billion. Take that part of the article with a salt lick
(and shame on Scientific American for not calling that out) and look to
the parts where it describes the specific circumstances needed to force
a collision.

Still, the closure contains good advice. When a chain has only one link
you know exactly where to look for that weak link. :)


> Meanwhile, as voice synthesis improves in fidelity...

That's only part of it. Even if (when) you can accurately reproduce any
arbitrary person's voice there is still the playback mechanisms. Voice
coils are mechanically incapable of exactly reproducing sounds. Even the
best speakers can be identified as speakers if your hearing is sensitive
enough and you know what to listen for.

On the flip side of that, if you inject the synthesized data stream
directly into system, bypassing speaker and microphone, it can still be
detected as a fake because it will lack the analog distortion expected
from the handset mic.

-- 
Rich P.