MSSpeech-Forum Homepage
Forum Home Forum Home > Windows™ Speech Recognition Forums > New Users & General Questions
  New Posts New Posts RSS Feed - Teaching WSR recursively
  FAQ FAQ  Forum Search   Events   Register Register  Login Login

Teaching WSR recursively

 Post Reply Post Reply
Author
Message
ed View Drop Down
Member
Member


Joined: 29/Nov/2009
Location: South of Boston
Status: Offline
Points: 18
Post Options Post Options   Thanks (0) Thanks(0)   Quote ed Quote  Post ReplyReply Direct Link To This Post Topic: Teaching WSR recursively
    Posted: 10/Mar/2014 at 4:32am
I'm wondering if my correcting Windows Speech Recognition recursively does more harm than good. An example:

Suppose I want WSR to render the phrase "my house." After a few iterations of the correction dialogue the closest I can get is "mine loaf." At this point is it wise for me to ask for a correction of "mine" (thus getting "my") and then to ask for a correction to "loaf" (thus getting "house"). Or does such a recursive procedure result in WSR taking longer than it otherwise would to interpret my speech properly? Thanks for any advice! - ed
Back to Top
mmarkoe_admin View Drop Down
Admin Group
Admin Group
Avatar

Joined: 16/Jul/2008
Status: Offline
Points: 331
Post Options Post Options   Thanks (0) Thanks(0)   Quote mmarkoe_admin Quote  Post ReplyReply Direct Link To This Post Posted: 10/Mar/2014 at 5:34pm
Originally posted by ed ed wrote:

I'm wondering if my correcting Windows Speech Recognition recursively does more harm than good. An example:Suppose I want WSR to render the phrase "my house." After a few iterations of the correction dialogue the closest I can get is "mine loaf." At this point is it wise for me to ask for a correction of "mine" (thus getting "my") and then to ask for a correction to "loaf" (thus getting "house"). Or does such a recursive procedure result in WSR taking longer than it otherwise would to interpret my speech properly? Thanks for any advice! - ed
Was my house just an example or something you have been having problems with? I'm going to try dictating your example. My house is on evergreen street. When people come to visit my house they are welcomed warmly.

First to answer your question. Speech recognition works not only by listening for the sound of each word, but for the context within which words are used. For example, they're going to park their car over there. Two boys went to see a Dr. because they eat too much food. The homonyms in these examples are recognized the cause of the context of surrounding words. Therefore, the best method for correction is to highlight two or more words and correct in that context.

Are you having accuracy problems in general? If so, you may have a problem with your microphone-not having high acoustic quality or it is picking up background noise. You can test for this in several ways. Try making a sound recorder recording and playback through speakers or earphones. The sound should be crisp and clear with no static or background noise. If it is not, consider a good microphone made for speech recognition.

Also, although this may sound like a plug for my WSRtoolkit software, it does have three ways to improve accuracy. This includes being able to add words or groups of words, creating a selection which to ensure user profile for the acoustic properties of your individual voice, and a feature that scans your typical documents and learns how you use language.

Marty Markoe, MyMSSpeech.com
Back to Top
ed View Drop Down
Member
Member


Joined: 29/Nov/2009
Location: South of Boston
Status: Offline
Points: 18
Post Options Post Options   Thanks (0) Thanks(0)   Quote ed Quote  Post ReplyReply Direct Link To This Post Posted: 10/Mar/2014 at 9:29pm
Thanks for the informative reply, Marty. Because my rejoinder is long, I put it first, and quote the thread afterward.  Apologies if this adds to anyone's confusion.

I'm not having problems in general. My SuperbeamArray, connected to an Andrea PureAudio SA at a front USB Port, works fine as long as the folks in the adjoining apartment aren't throwing one another against the walls. I have WSRtoolkit 3. It's very impressive, and I've used it in previous rebuilds of my system. (Don't let me get started on rebuilds...)

Though I've installed toolkit 3 in the present build, I haven't yet much run WSR through it, since I haven't had the time to properly train WSR in it and replace the profile I've built up in WSR/WSRMacros
alone. My note was not motivated by the actual errors I'm getting as much as by a desire to understand the underlying algorithm(s) and how they respond to particular styles of asking the program for corrections. I see from your reply is that my short schematic example was rather unclear. I don't know how much it will help anyone reading this, but I'll go through my process in detail, stepping through your example sentence:

    When people come to visit my house they are welcomed warmly.

 I dictate and get:

    When people come to visit this in my house Teruel com warmly

I know that there are several ways to correct this. (1) Focusing on short phrases, I ask for a correction of "this in my house" (getting"my house" in two iterations of the correction dialogue). Next, I ask for a
correction of "Teruel com" (again getting "they are welcomed" in two iterations). Pretty good, I think. (2) But what I usually do - out of laziness, I suppose -  is ask for correction of the *whole incorrect sentence*. This often works quite well, getting me a completely correct sentence. But this time I got:

    When people come to visit my house their welcome warmly

So this is my question. If at this point I choose to correct "their" to " they are welcome," does this cause
the program to learn more slowly than if I employed the previous shorter-phrase method? Another way to ask this might be, is it better to correct longer or shorter phrases? Or is the length of the phrase
irrelevant?

Perhaps my question doesn't make much sense in terms of how the algorithms actually work. Whatever
the case, I'd appreciate your reaction. My apologies if I'm bothering you and the other readers with the
obvious. Thanks very much for your help! - ed

Originally posted by mmarkoe_admin mmarkoe_admin wrote:

Originally posted by ed ed wrote:

I'm wondering if my correcting Windows Speech Recognition recursively does more harm than good. An example:Suppose I want WSR to render the phrase "my house." After a few iterations of the correction dialogue the closest I can get is "mine loaf." At this point is it wise for me to ask for a correction of "mine" (thus getting "my") and then to ask for a correction to "loaf" (thus getting "house"). Or does such a recursive procedure result in WSR taking longer than it otherwise would to interpret my speech properly? Thanks for any advice! - ed
Was my house just an example or something you have been having problems with? I'm going to try dictating your example. My house is on evergreen street. When people come to visit my house they are welcomed warmly.

First to answer your question. Speech recognition works not only by listening for the sound of each word, but for the context within which words are used. For example, they're going to park their car over there. Two boys went to see a Dr. because they eat too much food. The homonyms in these examples are recognized the cause of the context of surrounding words. Therefore, the best method for correction is to highlight two or more words and correct in that context.

Are you having accuracy problems in general? If so, you may have a problem with your microphone-not having high acoustic quality or it is picking up background noise. You can test for this in several ways. Try making a sound recorder recording and playback through speakers or earphones. The sound should be crisp and clear with no static or background noise. If it is not, consider a good microphone made for speech recognition.

Also, although this may sound like a plug for my WSRtoolkit software, it does have three ways to improve accuracy. This includes being able to add words or groups of words, creating a selection which to ensure user profile for the acoustic properties of your individual voice, and a feature that scans your typical documents and learns how you use language.

Marty Markoe, MyMSSpeech.com
Back to Top
mmarkoe_admin View Drop Down
Admin Group
Admin Group
Avatar

Joined: 16/Jul/2008
Status: Offline
Points: 331
Post Options Post Options   Thanks (0) Thanks(0)   Quote mmarkoe_admin Quote  Post ReplyReply Direct Link To This Post Posted: 11/Mar/2014 at 4:55am
Originally posted by ed ed wrote:

So this is my question. If at this point I choose to correct "their" to " they are welcome," does this cause
the program to learn more slowly than if I employed the previous shorter-phrase method? Another way to ask this might be, is it better to correct longer or shorter phrases? Or is the length of the phrase
irrelevant?

WSR uses bigrams, trigrams and quadgrams to analyze groups of 2,3, or 4 words at a time. Therefore, it learns based on usage within phrases of 2,3, or 4 words not whole sentences.
Marty Markoe, MyMSSpeech.com
Back to Top
ed View Drop Down
Member
Member


Joined: 29/Nov/2009
Location: South of Boston
Status: Offline
Points: 18
Post Options Post Options   Thanks (0) Thanks(0)   Quote ed Quote  Post ReplyReply Direct Link To This Post Posted: 11/Mar/2014 at 7:15am

Originally posted by mmarkoe_admin mmarkoe_admin wrote:

Originally posted by ed ed wrote:

So this is my question. If at this point I choose to correct "their" to " they are welcome," does this cause
the program to learn more slowly than if I employed the previous shorter-phrase method? Another way to ask this might be, is it better to correct longer or shorter phrases? Or is the length of the phrase
irrelevant?

WSR uses bigrams, trigrams and quadgrams to analyze groups of 2,3, or 4 words at a time for context. Therefore, it learns based on usage within phrases of 2,3, or 4 words not whole sentences.
Marty Markoe, MyMSSpeech.com
OK, thanks!

Edited by mmarkoe_admin - 11/Mar/2014 at 7:28am
Back to Top
 Post Reply Post Reply
  Share Topic   

Forum Jump Forum Permissions View Drop Down

Forum Software by Web Wiz Forums® version 12.02
Copyright ©2001-2019 Web Wiz Ltd.

This page was generated in 0.313 seconds.

Microsoft Most Valuable Professional

§- Thank you for visiting our Windows Speech Recognition and Macro Forum.. -§