From: Ed Kummel (tech_ed_at_yahoo.com)
Date: Tue Aug 16 2005 - 09:07:13 PDT
I believe that it has quite a bit to do with
horsepower.
The best voice recognition systems are not a single
system. They combine voice recognition, with word
identity and apply context using gramatical rules.
This can take quite a bit of horsepower. Back in the
early 80's, I had voice recognition on an old Radio
Shack color computer. It was a true voice pattern
recognition in that it only had the space for 300
words, you would type the word, then say the word 3
times, these three voice patterns would be stored for
the word (or some kind of hash combining all three
voice patterns...I don't know which) and you had to
use distinct speach, disassociated words, not fluid
sentences. It was almost 100% accurate and worked very
well...it was just hard on your vocal chords to talk
that way.
I played around with various voice recognition
systems, including lernhout and haughspie and dragon
systems. I even helped set up an Audix voicemail
system many years ago. The Audix worked the best
because you only had a limited number of responses for
any given voicemail prompt.
I stopped playing with this stuff many years ago...no
compelling reason for it.
Then, a recent broken hand brought it back to my
attention. I coupled Microsoft's built in Voice
Recognition in Office, with a new Bluetooth headset,
and I found that after 15 minutes of training, I was
getting better than 90% recognition for letter and
email writing! Nagivation is a different story, but I
believe that that too could be worked out. If I
modified my speech slightly, I could get even better
recognition! In fact, there were several postings I
made to NewtonTalk using the Voice Recognition
software.
The biggest drawback I found is a non intuitive method
of editing my recognized text. Switching between
command and text mode was a pain.
This was using a Dell d600 laptop running a 1.6ghz P4
and 512mb ram, and I was frequently getting pauses in
recognition that would require me to repeat the word
because the computer was hung on a previous
recognition. I believe that a less powerful machine
would make voice recognition worthless.
The biggest issue with voice recognition is the lack
of a killer app. For non-handicapped individuals,
there is no real compelling reason to use voice
recognition as a general use.
But in my old house (still haven't unboxed everything
after moving into my new house since February) I used
rudimentary voice recognition in rooms to control
certain devices. This consisted of simple phrases like
"living room lights on" and such. The biggest pain was
the programming and the wiring for the microphones...I
particularly enjoyed the "all lights on" command. From
any room, I could turn on all the lights inside and
outside the house. Too bad I never had a burgler to
try this out on!.
Ed
web/gadget guru
Download Newton packages directly onto your Newton:
http://newton.tek-ed.com
--- Jon Glass <jonglass_at_usa.net> wrote:
>
> On Aug 16, 2005, at 5:54 AM, Brian G. wrote:
>
> > If your typical cell phone can voice dial, why
> can't we have simple
> > voice commands for common tasks?
>
> That's because that isn't real voice recognition.
> It's pattern=20
> matching, and only rudimentary. I have about a 75%
> accuracy rate, so I=20=
>
> gave up. I guess my voice is too fluid or
> something...
> >
> --=20
> -Jon Glass
> Krakow, Poland
>
> "If your pictures aren=92t good enough, you=92re not
> close enough." - =20=
>
> Photojournalist Robert Capa
"I'm not an expert, but I *did* stay at a Holiday Inn Express once..."
--Holiday Inn Commercial
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com
-- This is the NewtonTalk list - http://www.newtontalk.net/ for all inquiries Official Newton FAQ: http://www.chuma.org/newton/faq/ WikiWikiNewt for all kinds of articles: http://tools.unna.org/wikiwikinewt/
This archive was generated by hypermail 2.1.5 : Tue Aug 16 2005 - 09:30:00 PDT