Registering and logging in removes this ad.
Registering and logging in removes this ad.
Sennheiser ME3 mike
Submitted by Dan Lufkin on Wed, 03/08/2006 - 22:40.
I finally faced the reality that my good old Andrea mike was getting noisy and ordered a Sennheiser ME3 with USB interface. WOW! WHAT A DIFFERENCE! I usually cruise along translating at about a thousand words an hour. I just finished a piece of 2700+ words in just over 80 minutes with a total of about three misrecognitions. I'd like the mike to be positioned to my left, but I guess I can fiddle with it a little. I get a s/n ratio of 25.
Dan
»
- Dan Lufkin's blog
- Login or register to post comments
Dan Lufkin wrote:I finally
I finally faced the reality that my good old Andrea mike was getting noisy and ordered a Sennheiser ME3 with USB interface. WOW! WHAT A DIFFERENCE! I usually cruise along translating at about a thousand words an hour. I just finished a piece of 2700+ words in just over 80 minutes with a total of about three misrecognitions. I'd like the mike to be positioned to my left, but I guess I can fiddle with it a little. I get a s/n ratio of 25.
Dan
May we quote you on our web site?
Marty
Marty -- Absolutely. I'm
Marty -- Absolutely. I'm pleased and a little surprised that the Sennheiser makes such a difference. I should mention that I just plugged it in the USB port and started talking. Now I'll take a little time off and train it right.
Dan
Dan, could you elaborate a
Dan, could you elaborate a little on what you mean by train it right?
Thanks, Jim Green
JimGreen wrote:Dan, could
Dan, could you elaborate a little on what you mean by train it right?
Frankly, I hope he does not do anything considering the accuracy he's getting. If it is not broken, do not fix it.
Probably what Dan was referring to was that we highly recommend you create a new user profile with a new better microphone. This is so it takes advantage of the better audio input. However, if someone hats exceptionally good enunciation and pronunciation it may not matter.
--
Martin Markoe, eMicrophones, Inc.
The best microphones for Speech Recognition
See us at: http://www.eMicrophones.com/index.asp
Read, "Key Steps to High Speech Recognition Accuracy" at:
http://www.emicrophones.com/docDetails.asp?DocumentID=38
I certainly agree with the
I certainly agree with the if it's not broke don't fix it approach. The reason I asked is I have changed my microphone settings in Windows but cannot find how to do the training or the microphone volume setting again from the Dragon menu. I guess I've been away from voice recognition too long to remember much.
JimGreen wrote:I certainly
I certainly agree with the if it's not broke don't fix it approach. The reason I asked is I have changed my microphone settings in Windows but cannot find how to do the training or the microphone volume setting again from the Dragon menu. I guess I've been away from voice recognition too long to remember much.
Click on the Tools menu, Accuracy Center, Check Your Audio Settings.
--
Martin Markoe, eMicrophones, Inc.
The best microphones for Speech Recognition
See us at: http://www.eMicrophones.com/index.asp
Read, "Key Steps to High Speech Recognition Accuracy" at:
http://www.emicrophones.com/docDetails.asp?DocumentID=38
Jim, As Martin says, it's
Jim,
As Martin says, it's usual to retrain DNS when you change anything in the audio. You just create a new user and do the regular training routine. My experience with the Sennheiser continues to be excellent, with a s/n ratio of 23-25, so I'm in no hurry to fool with it. "The perfect is the enemy of the pretty good."
I'm a linguist by trade and used to work as a radio announcer so my diction is probably better than average. My favorite DNS demo sentence is "Amidst the mists and coldest frosts, with barest wrists and stoutest boasts, she thrusts her fists against the posts and still insists she sees the ghosts." If you can dictate that, you can dictate anything.
Dan
Ye Gads! If you get 100% I
Ye Gads! If you get 100% I have tremendous respect for your announcing skills!
I got on first try:
"Amidst the mists and coldest frosts, Brest wrists and status boasts, she thrusts her fists against the post and still insist she sees the ghost's."
After corrections, the second try is perfect, including all plurals. I thought I said "with barest", but playback gave no indication of the "with", so I probably got so wrought up in the task I just skipped it. Of course, enunciating those "sts" is an unnatural act requiring my utmost effort.
Can't explain why I got "Brest" instead of "breast" since I say them identically, but SFAIK I've never had occasion to dictate the former!
Bruce
Thanks for all the replies.
Thanks for all the replies. I think I understand now. You just create a new user and that let's you do the retraining with a new or changed microphone set up. Today I bought a new microphone (the RadioShack knockoff) and USB pod from Martin, so I'll wait until that arrives and then retrain. But what amazed me is the accuracy I'm now getting using my six to seven year old Andrea microphone plugged into the sound chip on my motherboard. Poor input hardware, but it still works very well.
BTW, I don't think I mentioned it, but one reason for my renewed interest in voice recognition is I just had surgery on my right hand.
Thanks again, Jim Green
I am always interested in
I am always interested in finding more passages which can be used to help to judge dictation accuracy. I had very good results with the famous Rainbow passage (99%). With the passage quoted in the previous post, I had only one mistake, a very curious one. I got varus instead of barest. I can't understand why. I did not even know that the word varus existed.
I use DNS 7 Preferred, with a Buddy Gooseneck desktop Microphone. My computer has a 2.6 Intel Celeron processor, 512 RAM, and an 80 GB hard disk (7200), with Windows XP Home edition. I work almost exclusively with Internet Explorer 6, Outlook Express 6 and Microsoft Word 2002. So I have hardly got the latest most sophisticated version of DNS.
Graham Runnalls wrote:I am
I am always interested in finding more passages which can be used to help to judge dictation accuracy. I had very good results with the famous Rainbow passage (99%). With the passage quoted in the previous post, I had only one mistake, a very curious one. I got varus instead of barest. I can't understand why. I did not even know that the word varus existed.
Yep, it does:
varus - A deformity in which the foot (or other body part) is turned inward.
Next stop, the Twilight Zone
Chuck
The rainbow passage is a
The rainbow passage is a good test for SR developers because putatively it samples all of the sounds of the English language. Therefore it is a good test for speakers to make sure they can enunciate all these sounds.
If a user has trouble with recognition accuracy while dictating the rainbow passage, the proper inference is that s/he has trouble speaking English, not that the SR product or the user's files are deficient. But most dictators get inflated accuracy figures for the rainbow passage, i.e., it overestimates normal accuracy, simply because it has so many distinct sounds.
There is also the "rainbow" effect, i.e., because people only dictate it when they are trying to test their accuracy, they tend to enunciate far more precisely and to speak at more measured rates than they use in normal dictation! In normal speech, there are so many homonyms and people slur so much that accuracy in practice drops below what is observed in the rainbow passage.
Also note that from the perspective of accuracy all marketing variations of a particular DNS version have essentially the same recognition engine, i.e., there is no accuracy difference between Standard, Preferred, Pro etc., except possibly for user additions like macros, customized vocabularies, etc.
Note, however, most users think 8 is superior to 7.
FWIW, your gear seems sufficient to me for high accuracy, although speed could flag a bit if you keep many applications open simultaneously -- a 50% or 100% increase in memory would help for that.
Bruce
BruceCyr wrote:The rainbow
The rainbow passage is a good test for SR developers because putatively it samples all of the sounds of the English language. Therefore it is a good test for speakers to make sure they can enunciate all these sounds.
If a user has trouble with recognition accuracy while dictating the rainbow passage, the proper inference is that s/he has trouble speaking English, not that the SR product or the user's files are deficient. But most dictators get inflated accuracy figures for the rainbow passage, i.e., it overestimates normal accuracy, simply because it has so many distinct sounds.
There is also the "rainbow" effect, i.e., because people only dictate it when they are trying to test their accuracy, they tend to enunciate far more precisely and to speak at more measured rates than they use in normal dictation! In normal speech, there are so many homonyms and people slur so much that accuracy in practice drops below what is observed in the rainbow passage.
Also note that from the perspective of accuracy all marketing variations of a particular DNS version have essentially the same recognition engine, i.e., there is no accuracy difference between Standard, Preferred, Pro etc., except possibly for user additions like macros, customized vocabularies, etc.
Note, however, most users think 8 is superior to 7.
FWIW, your gear seems sufficient to me for high accuracy, although speed could flag a bit if you keep many applications open simultaneously -- a 50% or 100% increase in memory would help for that.
Bruce
Bruce,
Kudos. You have hit the nail on the head for at least a core of the problems with accuracy and with user corruption. I need to think about it, but I may post some common reasons why what you are saying is so critical to good (optimal) accuracy.
Bottom line is that whatever approach any one user takes to keeping their users in good shape, the key is how well you enunciate and remembering that we all seldom say the same thing the same way twice. For example, training words or phrases out of the context of normal dictation almost invariably results in our speaking words and phrases differently. This is why general training is more effective than training during correction. When you correct a misrecognition and listen to how you enunciated the original dictation, either leave it alone, or concentrate on speaking it the same whay that you did when you originally dictated it. However, there is more to it than just this and this is only one example of how user files can change for the better or for the worse.
Chuck
Chuck Runquist wrote: Bottom
Bottom line is that whatever approach any one user takes to keeping their users in good shape, the key is how well you enunciate and remembering that we all seldom say the same thing the same way twice.
we emphasize several things and training users. First of all, we highly recommend running the audio setup wizard at the beginning of a dictation session. When accuracy seems to fall as it does for all of us, we recommend running the audio setup wizard again. Besides adjusting for variations that occur in one's voice and in background noise, it puts us into what we call dictation mode. Which is just as you say, enunciate and speaking phrases and you get high accuracy.
--
Martin Markoe, eMicrophones, Inc.
The best microphones for Speech Recognition
See us at: http://www.eMicrophones.com/index.asp
Read, "Key Steps to High Speech Recognition Accuracy" at:
http://www.emicrophones.com/docDetails.asp?DocumentID=38
I used to think your daily
I used to think your daily audio wizard schtick was a symptom of fuddy-duddyism, but nowadays I tend to think its not such a bad idea. Now if I could just work into the shower/shave/hair combing/teeth brushing/dressing/etc. routine
Bruce
You mean I qualify as a sub
You mean I qualify as a sub for one of the twelve monkeys randomly dictating text in the hopes that a Supreme Editor can extract a nugget or two of wisdom
Bruce