[Discuss] Fidelity voice-recognition security?

[rlk at alum.mit.edu (Robert Krawitz)]

On Wed, 22 Nov 2017 17:55:32 -0500, Richard Pieri wrote:
> On 11/22/2017 1:44 PM, Robert Krawitz wrote:
>> And voices do change, both short and long term.  What happens with
>> voice ID when you have a respiratory infection, blocked sinuses, what
>> have you?
>
> Which is why any voice authentication system needs some leeway in
> matching attempts with the baseline. And of course it needs to adapt to
> individuals' vocal changes over time.

With that much leeway, there's more chance for collision, right?

> Voices can be recorded but this isn't necessarily good enough. Then
> again, POTS is restricted to 300Hz to 3kHz, and any system intended to
> operate in this range is going to have problems. But this isn't a
> problem intrinsic to voice authentication in principle; it's a flaw in
> these specific instances. Then again, again, the number of potential
> users limited by POTS restrictions is dwindling. Do you use any kind of
> voice over digital network like LTE or digital cable or FTTP or WiFi
> calling? Do you use standalone VoIP or chat applications? If so then
> you're getting 50Hz to 7kHz or better which is more than enough to
> capture low and high frequency harmonics needed for accurate voice
> authentication.

Sometimes LTE.  VOIP systems have too many dropouts to be very useful.

But I'd really like to see evidence (in the form of, say, suitable
peer reviewed papers) that individual voices are that unique and
identifiable.
-- 
Robert Krawitz                                     <rlk at alum.mit.edu>

***  MIT Engineers   A Proud Tradition   http://mitathletics.com  ***
Member of the League for Programming Freedom  --  http://ProgFree.org
Project lead for Gutenprint   --    http://gimp-print.sourceforge.net

"Linux doesn't dictate how I work, I dictate how Linux works."
--Eric Crampton