Does WSR really learn when corrected?
Printed From: MSSpeech-Forum
Category: Windows™ Speech Recognition Forums
Forum Name: New Users & General Questions
Forum Description: Ask questions, give and get answers.
URL: https://www.msspeech-forum.com/forum_posts.asp?TID=163
Printed Date: 27/Apr/2024 at 5:02am
Software Version: Web Wiz Forums 12.02 - http://www.webwizforums.com
Topic: Does WSR really learn when corrected?
Posted By: bipolarbear
Subject: Does WSR really learn when corrected?
Date Posted: 03/Feb/2010 at 8:03am
|
I was wondering if it really does learn every time you correct and spell a word it incorrectly interprets. I know the documentation says it does, but so does dragon say that, and I've personally corrected the same mistake literally hundreds of times in dragon and they persisted, particularly in version nine but version 10 as well.
I've already seen with WSR that custom words added to the dictionary are still in correctly recognized even after having been corrected and spelled a dozen times. Thanks
|
Replies:
Posted By: mmarkoe_admin
Date Posted: 03/Feb/2010 at 10:21am
The trick with any speech recognition software is to not correct single words. Always correct words in the context and the probability of the word appearing correctly the next time you use it in that context is very high. For example, if I say, "I wish to free slaves in sub Saharan Africa," and it comes up with, "I wish to feed slaves...." I simply say, Correct "to feed slaves" and make the correction, "to free slaves."
Marty
|
bipolarbear wrote:Posted By: russ
Date Posted: 03/Feb/2010 at 2:21pm
|
bipolarbear,
Just to add to what Marty is saying...
In addition to adding words to the dictionary, you may want to try recording pronunciations for the words that are giving you problems. Also, you may want to use the "prevent the word from being dictated" option if it's appropriate. This mostly works well for obscure words that you don't usually dictate. For example, I have a friend named Arne; this name is usually spelled Arnie, so I have prevented the common spelling from being dictated, and added the less common spelling to my speech dictionary.
-Russ
|
Posted By: bipolarbear
Date Posted: 04/Feb/2010 at 1:34am
|
I always record and train whenever the option is not listed in the correction menu. I also correct more than just the one word to lend context, when possible. But I've noticed with both programs, it doesn't seem to make a difference. Well that's not entirely accurate, occasionally it does, but only occasionally.
In dns, I've even exaggerated pronunciations to differentiate them from one another even more, and that hasn't helped a bit. It will continually make the same mistake. One improvement in version 10 over version nine, at least it doesn't begin sentences with "know", instead of "no". I don't know how many times I trained it in version nine, and it continually made the same error.
"Prevent from being dictated" is a terrific idea. It has limited use but in those situations where it was applicable, it solved my issues. Unfortunately it's not applicable terribly often.
|
Posted By: himAGAIN
Date Posted: 23/Nov/2010 at 4:56am
Helloooo Marty, THIS advice of yours should be in coloured capitals in the introduction to the whole idea of S2T (speech-2-text)! Most users make the same few simple mistakes: 1. Don't speak in meaningful bites - from comma to comma, or full stop! 2. Speak too slowly or too quickly. One should read as a BBC (England) Newsreader. 3. Fascinatedly watch/wait for the text to display on screen. Turn it off! Don't watch! 4. DON'T try to correct by using a different voice. You'll forget and say "wanna do it" next time, again. 5. Before you get mad at the program - test your typing speed AND to the corrected text time in w.p.m. I, of course don't do any of these things ..........  Cheers! HimAGAIN |