Re: Using voice/speech-recognition software

Subject: Re: Using voice/speech-recognition software
From: P Newman <pnewman1 -at- HOME -dot- COM>
Date: Fri, 6 Aug 1999 08:15:46 -0400

I had replied to the original posting offline. Just some comments.
Purely my own observations. I tested both Dragon and L&H on a PII 400
with 128 MG.
L&H has a initial five minute (training) enrollment period.
NaturallySpeaking takes about fifteen minutes. With both, you must
continually correct errors. Both claim to integrate seamlessly with
Word and WordPerfect, although there is a problem with jumping cursors
in WordPerfect8. For command and control, L & H has a slight edge. For
context recognition NaturallySpeaking has a very large edge. Try the
following as a test of context recognition: "Mr. Wright was right when
wanted to write about the time he went to see the beautiful blue sea";
"Which witch is which." "The Count of the County of Queens cannot
count" All of these will require some minor training to get the
correct capitalization, but in learning NatSpeak seems to have a
slight edge. If you use the word processing program that is supplied
with each program, both have excellent and easy correction, however,
the Dragon product seems to correct itself easier. L&H support seems
slightly more responsive. On balance, I personally prefer using
NaturallySpeaking, despite the jumping cursor problem in WP8. I had
tried the Phillips product and it just did not hold a candle to the
others. since VV did not send me an NFR product, I have no experience
with it.

HTH
Peter

----- Original Message -----
From: Candace Bamber <cbamber -at- CASTEK -dot- COM>
To: <TECHWR-L -at- LISTSERV -dot- OKSTATE -dot- EDU>
Sent: Friday, August 06, 1999 8:01 AM
Subject: Re: Using voice/speech-recognition software

> Hi Lori,
>
> I use IBM Via Voice, Executive Edition and have been doing so for
about a year.
> In fact, I am dictating this message. I am using a Pentium 266 -
which is the
> slowest machine I imagine you could use. And I take a giant
performance hit when
> I dictate directly to Word or other applications (like this Lotus
Notes Email
> program, for example). But, I had a goal to cut my keyboarding by
30-40% and
> that has certainly happened, along with a huge improvement in my
hands, wrists
> and shoulders (yes, the cure for RSI is to avoid the repetitive
movement that
> causes the stress injury!)
>
> Here are my answers to your questions:
>
>
> >1. What kind of work are you doing with it (e.g.,
> > primarily translation, manual composition,
> > white papers)?
>
> I do all kinds of text entry with it - manual writing, status
reports,
> proposals, RFI responses, lots of various reports, project plans,
email etc. I
> think what I don't do may be of more interest: I don't edit with it.
That is, I
> don't enter updates by voice and I don't do on-line review of other
people's
> materials with it. It's just too cumbersome. But for strict text
entry, it can
> "type" about 120 words a minute, which is a lot faster than I can,
and it's a
> much better speller :^)
>
> >2. How long is the startup?
> It takes a couple hours to set up if you're thorough. You read a
passage to it
> that takes about 40 minutes. Then you feed it samples of your
writing until
> you're bored senseless. It recognizes speech by "listening" to your
sounds, and
> matching what it "hears" against a list of the words you were most
likely to
> have used, based on the rules of grammar and your writing style,
which it
> analyses from your samples.
>
> Then it takes about another 6 weeks of using the "correction window"
to correct
> any mistakes it makes. When you use the "correction window" it
"learns" the
> proper word. If you don't use the correction window, it doesn't
learn the word
> and you keep correcting that word manually forever. This is
especially useful
> for acronyms and industry-specific words, but it requires a lot of
patience!
>
> >3. How accurate is it? Of 100 words, how many
> > do you have to correct? What about punctuation?
>
> I'm actually not a very accurate or fast typist, so I found the
accuracy and
> speed to be an improvement. If I'm dictating work-related stuff it's
very
> accurate because VV is very well trained on our house style and
vocabulary. I
> usually have to make corrections to emails because it is not as well
trained to
> my "personal" voice. I will forgo correcting this mail so you can
see for
> yourself how accurate it is. Punctuation is actually not a separate
issue. What
> I actually say is: Punctuation is not a separate issue PERIOD
Instead COMMA you
> name the punctuation as you go along PERIOD
>
> >4. Does environmental noise affect the software? We
> > are in an open cubicle environment (and I keep having
> > nightmares about that Dilbert cartoon where someone
> > looks over Dilbert's shoulder and yells, "Close!").
>
> Sometimes funny things happen when it picks up unusual ambient
noises, but if
> you train it in a noisy environment, it adjusts.
>
> >5. What are the optimal system requirements?
>
> I have a pentium 266.I can't imagine trying to run it on anything
slower. I'm in
> line for a 400, and I'm dying to get it!
>
> You didn't ask for other issues, but I'll give you two:
> If you are running VV with WORD, your personal voice file is very
vulnerable to
> Word crashes and file corruptions. You must save a copy of your
personal voice
> file to another directory every day. Also, you cannot install VV and
Robohelp or
> Doc To Help on the same machine. The DLL's and registry changes at
install
> interfere with each other Very bad.
>
> Another key issue is training yourself to use the software. Language
to be
> spoken and language to be written are processed by different parts
of the brain.
> It's a very difficult adjustment to learn how to write with your
mouth rather
> than your hands. As well, you have to learn all the commands. I am
now at the
> stage where I am developing "dictation" macros for a lot of the
things I say
> often (for example, if I say ME WORK, my sig line displays as at the
end of this
> email, or if I say CASTEK SHORT, a three paragraph marketing text
about Castek
> displays). As I add more macros, the language I use for dictation
becomes more
> cryptic and personalized as well as more efficient. It takes a lot
of practice.
>
> In fact, the whole thing is a lot of work and requires tremendous
patience. In
> my case it was worth it because my other option is to not work -
which isn't
> really an option.
>
> I hope this is helpful. If you have any further questions you think
I can help
> with, feel free to email me.
>
> (ME WORK :^)
>
> Candace
> ******************************
> Candace Bamber
> cbamber -at- castek -dot- com
>
> Castek
> --Putting the Future Together
> ******************************
>
>
======================================================================
=====
> Send commands to listserv -at- listserv -dot- okstate -dot- edu (e.g., SIGNOFF
TECHWR-L)
>
>
>

From ??? -at- ??? Sun Jan 00 00:00:00 0000=


Previous by Author: Re: TW impaired CW?
Next by Author: Re: voice recognition (cont'd) and IBM
Previous by Thread: Re: Using voice/speech-recognition software
Next by Thread: Re: Using voice/speech-recognition software


What this post helpful? Share it with friends and colleagues:


Sponsored Ads