Return to the archive index

Re: Universal Translator

Thanks Mark for the update, having read about it long ago, I neglected to follow these efforts
thinking it was stand allone, not the appearantly wise opportunity to add to the fuzzy sound
content in voice recogniton. (reading disabilities do take their toll).

A 1 to 17 probability would drop fast even with the fuzzy sounds, as a "b" and  a "p" (same lips)
should easily catch the voiced "b" vs "p" unvoiced.

It seems happily that there are many seeking to improve via lips in the (strill) not so great
state of the art in voice recognition.

Sorry to sprinkle a little water on Xunker's flames, but with a breakthrough in voice rec who
knows what will break through?,,, If so,, the vast # of voice apps then comming to life would
likely shrink fast to the low cost light weight ICs for those 'single chip' based WCs we'll be
wearing someday, hopefully long before mearly just surfing in a nursinghome.

Appearently an hmd though not required for face/lip finding, would still provide better targeting
(personal selection) and of course the very *many* other hmd values to the rigs.

Mark Billinghurst wrote:

> Alex Waibel at CMU has built computer vision and neural net based lip
> reading software. You can read about it at
> http://www.is.cs.cmu.edu/ISL.multimodal.lips.html
>
> There are several other groups doing very good work in vision based
> lipreading. These groups are mostly trying to use lipreading as another
> channel to improve speech recognition.
>
> See http://www.cs.berkeley.edu/~bregler/vaspeech.html for another site
> with more information..
>
>         Mark
>
> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> Mark Billinghurst               | Human Interface Technology Laboratory
>         | University of Washington, Box 352-142
> fax: +1-206-543-5380            | Seattle, WA 98195
> On Wed, 2 Dec 1998, Tristram W. Metcalfe III wrote:
>
> > Lets see, lipreading was, HAL 9000 two years from now,,
> >
> > The realities of reading lips are an enormous task of guessing up to 1 out of 17
> > possibilities per consonant or vowel X the real time rate of the phoneme stream. This
> > requires knowledge of the context and potential words, to begin closing the loop on what is
> > actually happening inside the mouth.
> >
> > When that soft ware is condensed down from its vapor, I agree it would be nice, maybe it's
> > really a project moving ahead somewhere, anyone know?.
> >
> > A thread many months ago touched on some interior mouth shape sensing and sending antenea,,
> > oops sorry,,  that only works on the user's mouth, not one across the room.
> > However, it certainly is worth exploring abilities to improve communication across the room
> > or table.
> >
> > Brenden Tuck wrote:
> >
> > > Hmm..this thread brings to mind an interesting experiment....
> > >
> > > I wonder how hard it be to teach a wearable how to "lip-read."
> > >
> > > Instead of using voice recognition, use a camera and
> > > modified face recognition algorythms.
> > >
> > > It would be more processor/hardware intensive, but
> > > I bet the data sets between different people using the same
> > > language wouldn't be as wildly different as their vocal patterns
> > > would be.  People tend to communicate facially in very similar ways.
> > >
> > > Facial hair would be a problem, most likely, but other things like
> > > lipstick would provide even higher contrast for the muscle patterns.
> > >
> > > --friar
> > >
> > > --
> > > Subcription/unsubscription/info requests: send e-mail with subject of
> > > "subscribe", "unsubscribe", or "info" to 
> > > Wear-Hard Mailing List Archive (searchable): http://wearables.ml.org
> >
> >
> >
> > --
> > ÐÏࡱ

--
ÐÏࡱá

--
Subcription/unsubscription/info requests: send e-mail with subject of
"subscribe", "unsubscribe", or "info" to 
Wear-Hard Mailing List Archive (searchable): http://wearables.ml.org

From Wear-Hard Mailing list Archive (WH)
Maintained by R. Paul McCarty

Archive created with babymail