MSSpeech-Forum Homepage
Forum Home Forum Home > Windows™ Speech Recognition Forums > WSRToolkit
  New Posts New Posts RSS Feed - Considerable problems with WSR stability
  FAQ FAQ  Forum Search   Events   Register Register  Login Login

Considerable problems with WSR stability

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


Joined: 22/Nov/2010
Location: Queensland, Aus
Status: Offline
Points: 7
Post Options Post Options   Thanks (0) Thanks(0)   Quote himAGAIN Quote  Post ReplyReply Direct Link To This Post Topic: Considerable problems with WSR stability
    Posted: 04/Dec/2010 at 11:29pm
Hi anyone...???
VERY quiet here.  I would have thought there would be a lot more activity. Especially with the new release.

OK - my problems:
1. Trying to train the program:
I took the option of pasting my own text into WSRT - recognition is all over the place due to extraneous noise sensitivity.(I think)
I have actually just started up the programme here in the form and I won't do any corrections or near Except in a new we put them in brackets and correct where I came stop

Oh ..... back to keyboard that was hopeless.!!!!
1. The training program would accept my pasting text into it, but as I tried to read it it would jump whole paragraphs - as I was reading it.
2. The accuracy will be quite good but only in dead silence - ANY external noise creates beeping and "What was that?" messages.  AND only in Wordpad - doesn't work with anything else - as shown above.
3. Training doesn't "take"  - I cannot train it to accept "full stop" as a Period (.).  Doesn't like "period" either, or "end-of-line" commands.

I have tried adjusting the sensitivity, but nothing seems to change - even if the ratios are accepted.

Really dislike the idea of being forced back to Dragon and MSWord.
Hoping someone can help with settings experience especially.
I am using an inexpensive desktop USB Logitech after wasting a lot of money on more expensive units over the years and it does reasonably well in Dragon - as good as any I've tried. Even in WSR so long as there is dead silence......Confused
 
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: 05/Dec/2010 at 9:02am

The following is the response sent to John Rigby's private E-mail:

Quote Something is seriously wrong with the training side in the toolkit and I don't think I would be alone
 

Are you mixing up the WSRToolkit (developed by us) with WSR developed by Microsoft? The WSRToolkit is just a tool. If there are flaws beyond our control we should not be blamed for them. I would start with looking at your microphone as no microphone used at a long distance will be as good as a good noise canceling close talk microphone. You said on the forum, “after wasting a lot of money on more expensive units over the years.” You don’t say what these more expensive units are? We could not find any record of purchases from us other than the WSRToolkit? There may be more expensive microphones that are not good for speech recognition.

 

Quote would I? I don't really understand the ratios of the sound setup too well, but nothing has worked above about 60% accuracy unless in dead silence.
 

We read your post. I can guarantee the problem is partly the microphone and also the audio subsystem of Vista and Windows 7. We don’t recommend any long distance from mouth microphone and especially not a Labtec.  If you pull out one of your expensive close talk microphones and set the Input volume to 20% and the Input SNR to 12 (as a starting point) you will see good results.

 

Marty Markoe, eMicrophones, Inc.



Edited by mmarkoe_admin - 05/Dec/2010 at 9:03am
Back to Top
himAGAIN View Drop Down
Member
Member


Joined: 22/Nov/2010
Location: Queensland, Aus
Status: Offline
Points: 7
Post Options Post Options   Thanks (0) Thanks(0)   Quote himAGAIN Quote  Post ReplyReply Direct Link To This Post Posted: 05/Dec/2010 at 5:01pm
Hello???
Have I been banned?
Just wasted an hour trying to post a reply here?   Twice.

Back to Top
himAGAIN View Drop Down
Member
Member


Joined: 22/Nov/2010
Location: Queensland, Aus
Status: Offline
Points: 7
Post Options Post Options   Thanks (0) Thanks(0)   Quote himAGAIN Quote  Post ReplyReply Direct Link To This Post Posted: 05/Dec/2010 at 5:50pm
Originally posted by himAGAIN himAGAIN wrote:

Hello???
Have I been banned?
Just wasted an hour trying to post a reply here?   Twice.



AAAAARRRRGGGGHHHH!!!!  
It did it again!  
Another half hour's worth! 

Back to Top
himAGAIN View Drop Down
Member
Member


Joined: 22/Nov/2010
Location: Queensland, Aus
Status: Offline
Points: 7
Post Options Post Options   Thanks (0) Thanks(0)   Quote himAGAIN Quote  Post ReplyReply Direct Link To This Post Posted: 05/Dec/2010 at 5:51pm
Hi Marty,
I just sent a test message as after spending an hour responding, your system wouldn't let me back in - "authority error" - as in usually being banned!!  But, here we are........

 
Quote Are you mixing up the WSRToolkit (developed by us) with WSR developed by Microsoft? The WSRToolkit is just a tool. If there are flaws beyond our control we should not be blamed for them. I would start with looking at your microphone as no microphone used at a long distance will be as good as a good noise canceling close talk microphone. You said on the forum, “after wasting a lot of money on more expensive units over the years.” You don’t say what these more expensive units are? We could not find any record of purchases from us other than the WSRToolkit? There may be more expensive microphones that are not good for speech recognition.

No,  the problems I now have are since I installed the WSRKit and actually relate to its specific abilities/services:
1. Changing sensitivity of mic.
2. Adding own text for training.
3. Type of mic has to be reset each startup.
4. 60% accuracy since the install.  I was originally impressed with WSR - it was equal to Dragon on a non-setup start.

I have used up to the "Rolls-Royce" Sennhauser (borrowed) and not been that impressed.  I don't like headsets and being tethered to the machine, and today, don't use wireless anything after reading genuine scientific reports on their dangers.

No, I hadn't purchased from you before - never heard of you - I'm in Oz...  :-)
I wasted a LOT of money over the years on equipment and for final value and application, the el cheapo Logitech amazed even me.  Certainly, the odd very expensive mics can produce a few percent more accuracy especially in difficult crowded/noisy situations, but at the cost of comfort and a lot of money.
For my needs as a writer/journalist the time factor is in thinking and composition and comfort is the biggest factor.
I mainly use portable dictation units, my old Sony SX70 producing best results of all equal to the more expensive Philips with DS modulation - around 96% accuracy overall in even non-ideal conditions on a beach.  (Not tested since installing the WSRKit and trying to change to WSR).

The Logitech for corrections and answering phones is fine - until now.



 
Back to Top
himAGAIN View Drop Down
Member
Member


Joined: 22/Nov/2010
Location: Queensland, Aus
Status: Offline
Points: 7
Post Options Post Options   Thanks (0) Thanks(0)   Quote himAGAIN Quote  Post ReplyReply Direct Link To This Post Posted: 05/Dec/2010 at 5:57pm
I think that your Server system is causing the problem by disconnecting if the subscriber spends more than "x" minutes composing a post.  
I've seen that before .............
Luckily I had copied most of my last post - after the experience today - and it went thru ok.

For now, I give up and will get to work - hope for a comment or two that can give me hope.

Cheers!

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: 05/Dec/2010 at 7:20pm
Originally posted by himAGAIN himAGAIN wrote:

I think that your Server system is causing the problem by disconnecting if the subscriber spends more than "x" minutes composing a post.  
I've seen that before .............
Luckily I had copied most of my last post - after the experience today - and it went thru ok.
This forum and our eMicrophones forum use the exact same software and reside on the same server. We've never heard of the forum software timing out. I'll leave this running while I go have dinner and then post. It is now 6:35 PM GMT -5
 
Also, I learned long ago when using Dragon to dictate to text fields in Internet Forums that it was dangerous to dictate directly. Too often, words dictated were interpreted as commands that took one out of the text window and when click back all text would be gone. For many years we've dictated into the Dragon Dictation Box and now with WSR we have a macro called OpenDictationBox.WSRMac that let's us dictate to Word Pad and transfer text to an Internet Forum text field.
 
Marty
 
PS. Finished dinner and posted this at 7:19.
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: 05/Dec/2010 at 7:31pm
Originally posted by himAGAIN himAGAIN wrote:

No,  the problems I now have are since I installed the WSRKit and actually relate to its specific abilities/services:
1. Changing sensitivity of mic.
2. Adding own text for training.
3. Type of mic has to be reset each startup.
4. 60% accuracy since the install.  I was originally impressed with WSR - it was equal to Dragon on a non-setup start.
Will respond to each of the above:
 
1. The audio subsystem problems with Windows Vista and 7 are well documented. Our Tune Microphone feature will override Windows. Number 3 in the Readme section:
VOLUME LEVEL ADJUSTMENTS
Once the values are changed in the profiles section, the WSR system should be restarted to accept the new values. BE ADVISED, if the 'Setup my Microphone' wizard is run, the values you set previously will be invalidated.
 
Have you followed the above?
 
2. How did you add your own text for training? What was in the text?
 
3. What do you mean the type of microphone has to be reset? Are you running the Setup my microphone utility?
 
4. If you believe WSRToolkit is the culprit with your reduced accuracy, simply uninstall it. Of course you used the Manage Profiles feature to backup your speech profile before playing with new features?
 
Marty Markoe
Back to Top
antihadron View Drop Down
Member
Member


Joined: 27/Feb/2009
Status: Offline
Points: 18
Post Options Post Options   Thanks (0) Thanks(0)   Quote antihadron Quote  Post ReplyReply Direct Link To This Post Posted: 19/Dec/2010 at 5:03pm
I've found that WSR is more sensitive to microphone quality than is DNS. I've tried to use cheapo headsets.  WSR picks up every bit of background noise while using them.   Also had problems completing training.  Commands tend to work OK, but dictation not so well.   Upgrade to a better Microphone :-)  Some are expensive http://www.emicrophones.com/p-1117-sennheiser-md-431ii-complete.aspx, some not so much. 
http://www.emicrophones.com/p-1199-samson-q7-dynamic.aspx

You mention you don't like to be 'tethered' to a headset.

 I use the wireless Samson Airline 77 http://www.emicrophones.com/p-1154-samson-airline-77-wireless-usb-headset.aspx.  It was expensive, but it works great with WSR. I think its one of the best ones out there. Very comfortable. Fantastic range.  Great accuracy.  Only negative is replacing batteries regularly.  This is mitigated with rechargeable ones.  Also wouldn't have minded a 'beep' when the battery is running low.  

I've also messed around with some of the bluetooth headsets (BlueAnt Q7) with mixed results.  WSR and bluetooth have some known accuracy/compatibility issues, but I found it moderately usable for commands and limited dictation. Your mileage may vary.
 
Back to Top
bipolarbear View Drop Down
Member
Member


Joined: 16/Jan/2010
Status: Offline
Points: 55
Post Options Post Options   Thanks (0) Thanks(0)   Quote bipolarbear Quote  Post ReplyReply Direct Link To This Post Posted: 24/Jan/2011 at 5:17pm
Actually it's that WSR has a much more sensitive intput level than dns. Microphone levels required to get good results from dns would absolutely pin WSR. They need two different intput levels
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: 24/Jan/2011 at 5:41pm
Originally posted by bipolarbear bipolarbear wrote:

Actually it's that WSR has a much more sensitive intput level than dns. Microphone levels required to get good results from dns would absolutely pin WSR. They need two different intput levels
You really need to reference the points you are referring to?
 
If you are referring to Bluetooth, you are incorrect. Bluetooth samples audio at 4 kHz except for the VXI Xpressway which has been reworked to sample audio around 9-10 kilohertz and thus be close enough to Dragon NaturallySpeaking's requirement of 11 kHz. WSR samples audio at 16 kHz and basically there's no Bluetooth that will work well with WSR because of the difference in sampling rates.
 
By the way, antihadron whose post you were responding to is the programmer who wrote the code for the WSRToolkit. He understands audio quite well as you can imagine. 
 
Marty
Back to Top
bipolarbear View Drop Down
Member
Member


Joined: 16/Jan/2010
Status: Offline
Points: 55
Post Options Post Options   Thanks (0) Thanks(0)   Quote bipolarbear Quote  Post ReplyReply Direct Link To This Post Posted: 24/Jan/2011 at 6:36pm
Actually, I'm referring to any microphone I've used with the two of them. In each case, optimal results with WSR required a much lower microphone volume than dns. This is empirically gathered information. Does not require any special audio knowledge or training, but even if it did, I think 24 years as a professional musician in live and studio situations would qualify me to speak on it.
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: 24/Jan/2011 at 6:39pm
Originally posted by bipolarbear bipolarbear wrote:

Actually, I'm referring to any microphone I've used with the two of them. In each case, optimal results with WSR required a much lower microphone volume than dns. This is empirically gathered information. Does not require any special audio knowledge or training, but even if it did, I think 24 years as a professional musician in live and studio situations would qualify me to speak on it.
Thank you for telling us what you were referring to. This is helpful.
 
antihadron wrote:
"I've found that WSR is more sensitive to microphone quality than is DNS. I've tried to use cheapo headsets.  WSR picks up every bit of background noise while using them.   Also had problems completing training. "
 
It sounds to me like everyone is on the same page. One of the programming requirements for the WSRToolkit Version 2 was for the Tune Microphone function which works great.
 
Marty
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 0.346 seconds.

Microsoft Most Valuable Professional

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