- Messages sorted by: [ date ][ thread ][ subject ][ author ]
- Next message: MDOTJOE@@aol.com: "[VRUG] My Experience with ver. 3. 01"
- Previous message: Aaron Green: "[VRUG] Consensus re: upgrade to NS 3?"
- In reply to: L.C.Jackson@@psd.maff.gov.uk: "Headset tip (was Mic Conundrum, New To List)"
- Next in thread: Martin Markoe: "Re: [VRUG] Thanks"
The notions that you espouse ----- 2+2=4 OR 2+2= [NOT 4] is far
too simplistic
to be of any use in an analysis of speech recognition performance.
If a speaker were to say the same word EXACTLY the same way
each and every time, this would be a different matter. However you
may NEVER say the same word exactly the same way twice but
want it understood as the same word every time, nevertheless.
There is fuzzy logic and there is _guessing_ and _best-guessing_
and taking the time to evaluate more possibilities and more
information and a larger knowledge base.
Give me a second to guess your height or give me an hour to
guess your height ---- see which guess will be better.
Both would be easy if I knew for sure that you were definitely either
six feet tall or three feet tall. If you can be any height between the
two, more time yeilds a better guess.
Peter M Lerman
(trying to sound like I know what I'm talking about)
Date sent: Wed, 12 Aug 1998 12:39:36 -0400
To: voice-users@@voicerecognition.com
From: Jim Green <jgreen@@robotvision.com>
Subject: Re: [VRUG] Accuracy vs. system speed?
> This may be a little off topic for this list, but may also address Rose's
> question about minimum system requirements for Nat 3.01. It has been
> suggested that Nat 3.01's accuracy may be better on "faster" machines.
> One advantage to dealing with computers is that they are dererministic
> devices, that is, if you add 2+2, you always get 4. (Humans are
> notoriously non-deterministic.) A faster CPU will get 4 faster, but the
> answer is always 4.
>
> The above suggestion amounts to saying that Nat 3.01 is **not**
> deterministic when run on a slow enough computer, that is, on a slower
> computer, 2+2 = 3.5!! If this is the case, then the threshold when Nat's
> output becomes non-deterministic is surely a "lower limit on system
> performance" to use with Nat.
>
> Regards, Jim Green
>
> --------------------------------------------------
> At 09:42 AM 8/12/98 -0500, Holland St.John wrote:
> >My comment about the reason for the "better accuracy with faster
> >processors" is my rehash of what I was told by one of the Dragon people
> >last year. I have not tried to determine accuracy differences with
> >different speed processors myself. My impression is that there are not
> >noticeable differences for me in accuracy in using Nat 1, 2, 3 on
> >"slower" and "faster" computers, but I have not tried the same program on
> >a Dragon spec minimum machine then compared it on a PentII-400 w 256 meg
> >ram.
> >
> >Holland
>
> #~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~#
> James E. Green, Ph.D. # Jim@@RobotVision.com # 703-757-5140
> Robot Vision Corporation
> Computer Vision System Design & Implementation
> 9978 Blackberry Lane # 6 Mountain Laurel Ranch
> Great Falls, VA 22066 # P. O. Box 569, Leakey, TX 78873
> http://www.next-wave.com/jim.htm
- Next message: MDOTJOE@@aol.com: "[VRUG] My Experience with ver. 3. 01"
- Previous message: Aaron Green: "[VRUG] Consensus re: upgrade to NS 3?"
- In reply to: L.C.Jackson@@psd.maff.gov.uk: "Headset tip (was Mic Conundrum, New To List)"
- Next in thread: Martin Markoe: "Re: [VRUG] Thanks"
 |  |