OT: Vista's "ultimate" security :)

[Fairlight]

Yo, homey, in case you don' be listenin', Bob Rasmussen done said:
> 1. Although Vista has just been released to retail, it has been available 
> to developers for many months. In fact, voice recognition has been part of 
> the TabletPC for years; presumably they have the same vulnerability, as 
> do Dragon, Via Voice, etc.

Voice Buddy.  I have that one ostensibly for gaming assistance but actually
worked out a profile whereby I could control PuTTY as well by speaking the
characters I wanted.  Also used to use it to trigger starting/stopping
playback in Media Player, and navigating various parts of Explorer.

> 2. I believe a software patch could prevent this, while still allowing 
> voice control. Such patch could operate similar to an anti-feedback check, 
> by preventing acceptance of audio that was currently being output.

That's going to be pretty processing-intensive, though.  If that were
-that- simple or lightweight, TeamSpeak and Ventrilo would already have
it.  That's worse than just checking for noise cancelling, you're trying to
catch specific signals, which are also going to be altered significantly by
the accoustics involved.  You're spitting a specific waveform out the DSP
of your sound card, but you're getting it back variably shifted depending
on the ambient noise, accoustical absorption of the environment, the
dynamic range and sensitivity of the mic at different frequencies (treble
and bass may fall off or become more prominent, independantly), and the
overdrive effect the speaker volume may have on the mic's interpretation
of what it's "hearing" (ie., in effect causing distortion sometimes equal
to a fuzz pedal on low)...and trying to reconcile the two waveforms.  Oh,
and by the way, trying to actually do it in "chunks" so that it can catch
discrete sounds and then handle them appropriately (ie., a game's explosion
isn't just one point in the waveform, it's a whole pattern of waveform
within the waveform that has to be isolated amongst the rest of the ambient
input completing the waveform, matched and cancelled out).  This is a -lot-
more complex than your basic background noise filters like those that come
with the Andrea mics that John introduced several of us to.  Those cancel
out -steady- background noise.  But acute noises are going to be far harder
to isolate, IMHO.  So you're almost inevitably talking about a significant
delay in responsiveness and expenditure of CPU--in an OS that's already
taking away from the applications' share of the CPU to begin with because
it's just plain more resource-hungry.

That's how I see that solution not panning out well.  

If a qualified sound engineer tells me I don't have the details right, I'll
take their word for it.  Bill Vermillion hasn't been subbed for ages, but
John Esak is still around and could indicate if I've got the right or wrong
ideas with regard to the aspects involved.  No idea who else here has done
significant audio work.  I feel I have just a firm enough grasp from work
with mixing, equalisation, and sampling to feel the above is pretty
accurate unless I'm contradicted by a knowledgeable specialist.

> 3. Most voice recognition programs have to be trained. It is unlikely that 
> a random voice command would be understood (although a low success rate 
> for a virus does not necessarily kill it).

I agree that most -seem- to have to be trained.  When I got Voice Buddy,
I had to install the MS Voice Recognition SDK for it to run.  It worked
pretty darn well out of the box.  I tried improving that with the MS SDK
training system.  After 10 sessions with it, it actually recognised my
speech -less- than when I started--by about 70% or worse degradation in
recognition.  I literally could hardly get anything to work anymore via
speech recognition.  So far, having since wiping the training to defaults,
my experience has taught me to leave it alone. :)

mark->
-- 
Try our new SPF-0 lotion, SunScream[tm].  Get it while it's hot!