Re: Massively parallel carbon-unit-based voice pattern matching

At 02:54 PM 8/9/96 -0700, Ernest Hua wrote:
>Now what if the FBI had voice detection systems just like the ones
>used in "Clear and Present Danger"?  Now THERE'S a good reason for
>tapping 1% of every phone switch ...
>
>I wonder if the public would mind having a MACHINE do drift net
>matching on voices on all public phone switches in the U.S. just to
>find the Olympic bomber ...  What if the specifications of that
>machine were made public and the machine placed under Congressional
>oversight?  (At least members of Congress would never get tapped.)

Voiceprints used to be doable only with sophisticated, specialized 
equipment.  Presumably, they would be doable today with "only" a PC and 
soundcard, along with some software.  This raises an interesting question:  
Would it be possible to modify speech to remove the identifiable 
characteristics which came from the speaker, and replace them with those of 
some other person.  

In other words, could somebody fake a bomb threat using a recording which 
has been processed to sound "exactly" like some famous person whose voice 
you can analyze?  


Jim Bell
jimbell@pacifier.com

> 
> In other words, could somebody fake a bomb threat using a recording which 
> has been processed to sound "exactly" like some famous person whose voice 
> you can analyze?  
> 

In theory, yes; with digitized sound, anything is "possible."

However, in practice, it isn't all that different from altering digitized 
photos. In theory, a "perfect" false image could be produced, but in 
practice, subtle errors are detectable. Once you've detected that 
alteration has occurred, the information is suspect, and the 
perpetrator has added clues as to his/her identity.

-r.w.

This CAN be infiltrated. (1) take into account line noise. (2) take into 
account recordings, cut and paste, and hell, audio programs are 
INCREDIBLE nowadays. (3) take into account problems like illness, which 
changes voice, inflection, etc., like puberty (hehe, no comment), like 
just about anything which requires some range or variability.

This seems to be about as secure, remotely, as NFS <G>.

Ta,
Millie, from my b/friends accout.
bug me, not him:   sfuze@tiac.net

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
                          SANDY SANDFORT
 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

C'punks,

On Fri, 9 Aug 1996, jim bell wrote:

> Would it be possible to modify speech to remove the identifiable 
> characteristics which came from the speaker, and replace them
> with those of some other person.  

Equipment exists that can remove the microtremors associated with
stress, thus defeating voice stress analysis technology.  Modeling
someone elses voice print would seem to be of little value, if
possible at all.

However, one could use voice-synthesis software to phone in a
bomb threat.  (The cops would probably arrest Stephen Hawkings 
on the basis of the voice print.)


 S a n d y

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Voiceprints can be done on a modern PC with a D/A sound input. The
software is, essentially, FFT, various kinds of digital filtering
and windowing, and display. Linguists (my "real" profession) have
a generally low opinion of voiceprint analysis.

For an excellent introduction to the technology, read Alexander
Solzhenitsyn's "First Circle." (The linguistics is accurate.)

Martin Minow
minow@apple.com