Question for Duncan Campbell re: Word-Spotting Capabilities

[Dave Bird]

In article <199907311227.IAA05607@smtp3.mindspring.com>, 
John Young <jya@pipeline.com> writes:
>One person wrote that he knew of a US company that boasted
>it had a product that could do several dozen simultaneous 
>translations on the fly, only available to the natsec realm.
>While this was initially greeted with skepticism (Duncan for one), 
>a bit of digging in the US Patent Office showed that there are well 
>over 300 patents utilizing an algorithm applied by the Hidden 
>Markov Model (HMM). It has great power to do pattern analysis 
>and is much in use by the voice and text recognition industry
>(though not limited to that).
>
>ATT has done a lot of research in voice recognition, primarily
>for public telephonic purposes, but the literature suggests that 
>classified work was (is being) done for the mil/gov. A classic
>primer in HMM was co-written by an ATT scientist, Rabiner,
>in 1986:



 Thanks for someone answering who has some real technical knowledge.

 As I suspected, to do any better requires targeting recognition on
 the next words and next phonemes most likely to follow what went 
 before (because this is how the brain does it); I don't know the
 specific maths.  I can't say much further, because I'm already
 out of my depth.


 

-- 
   ^-^-^-@@-^-;-^   http://www.xemu.demon.co.uk/
        (..)__u     news:alt.smoking.mooses