[comp.sys.dec] voice recognition, comerical products

davidd@wolf.cs.washington.edu (David Doll) (12/01/90)

If memory serves me right, I saw at the latest SIGGRAPH in Dallas, DEC's booth
had some voice recognition system running -> does anybody know anything about
this?
Also could you share any knowledge about researchers/products/etc dealing with
this issue...Thanks.

David

jg@crl.dec.com (Jim Gettys) (12/01/90)

In article <DAVIDD.90Nov30110513@wolf.cs.washington.edu>, davidd@wolf.cs.washington.edu (David Doll) writes:
> If memory serves me right, I saw at the latest SIGGRAPH in Dallas, DEC's booth
> had some voice recognition system running -> does anybody know anything about
> this?
> Also could you share any knowledge about researchers/products/etc dealing with
> this issue...Thanks.

The demo was called "diddly", done by Dave Wecker and Dick Helliwell of our
database organization.  The fun thing about the demo is that it isn't smoke
and mirrors (though one might believe it was held together by spit and baling
wire); the technology all really exists.

It used a bunch of technology, all of which already runs on our machines,
some of which is research tools.  Dave and Dick put it together in
six weeks.

It used a neural net to detect the keyword "diddly", (acronym "Digital's
Integrated Distributed Database Laboratory)which enabled 
the speech recognizer to parse the following phrases for commands to the system.
It uses Dave's neural net software;
I believe it is available for anonymous ftp from gatekeeper.
The point of this is that you can afford to have such a single keyword
recogniser running at all time, and do some useful work with the machine.

It used CMU's Sphinx continuous speech package for the voice recognition itself.
It used the DECvoice software for text to speech (running native on the DS5000, 
not via a DECtalk or DECvoice hardware; the DS5000 is fast enough to run it
directly, with no additional hardware).  
The imaging stuff is an assortment of publically available X tools,
a few simple internal imaging tools, the DECimage product, and other DDIF tools.
It used the software video stuff being done at Digital's Campus Engineering
Center in Karlsruhe, Germany.

The audio interface was a TURBOchannel card, built for a different purpose,
this interface will never see the light of day, but diddly will live on on
a much better audio interface.
Digital is working on a general audio capability for our multi-media work.
You'd have to talk to your salesman to find out details about it.
				- Jim Gettys