With help from my institution's techsupport, last summer I succeeded in getting Naturally Speaking Preferred 8.0 to run on a MacBookPro using Parallels & a Plantronics CS55 microphone.  At the time, the speech noise ratio was about 24, and I had reasonable accuracy.   Before I'd done much work with the set-up, however, the MacBookPro proved to be a lemon.    I am now trying to reconfigure Dragon on a replacement MBP, and cannot remember what the the previous Windows settings were.   Windows recognizes the mic on the "line in" setting,  but the speech noise ratio is about 31.  I've given up on completing training because its clear that s & f are going to present insurrmountable  accuracy problems.   Anyone have any suggestions?  (Aside from switching entirely to a Windows machine, which is not an option.)

We recommend using your ANC USB Pod external USB soundcard.  You may have had a driver issue with your previous computer but we doubt you'll experience the same problem with your new computer.  We recommend USB Pods for all notebook computers; including Macs.

You can plug your Hybrid Plantronics CS55 into the microphone input jack but it probably won't work because Macintosh systems require a longer plug (or at least they did a few years ago when we last checked).  We recommend connecting your CS55 to your USB Pod.

-------------------------

Click KB 2012 REV D to Download a 30 Day Evaluation of KnowBrainer 2012
Click SpeechStart+ to Download a 15 Day Evaluation of SpeechStart+

Lunis,

Unfortunately, Apple specializes in proprietary.  They do incredible work but now we’re all stuck with FireWire and USB.  FireWire is a better concept but it would be nice if Apple would simply play ball with everyone else.  Apparently they have agreed to play ball on the microphone issues so you can forget our previous advice about the microphone not working.  Apparently the CS55 is working in your system but keep in mind that you are still using an integrated soundcard so the results may not be quite as good as with an external USB sound card.

Volume could definitely be a factor.  The blue bar, in the first stage of the Audio Setup Wizard, should land between 1/3 to 2/3 of the way down.  If the blue line is close to or on the bottom then you have too much volume.  If you are experiencing too much volume, you may want to contact us for a $8 attenuator (volume control).

-------------------------

Click KB 2012 REV D to Download a 30 Day Evaluation of KnowBrainer 2012
Click SpeechStart+ to Download a 15 Day Evaluation of SpeechStart+

Lunis - I was not aware that a speech noise ratio reading of 31 was "too high."  What do the ranges mean?  I.e., what is the best range to get:  24-26?  I know that 16-18 is not so good (accuracy not too good), but I was always under the impression that higher was better.  No?

 

-------------------------

Dana Joan - Vero Beach, FL  - 

Dragon NaturallySpeaking Medical, Version 10.1; General Medical Large Vocabulary; Windows 7 Ultimate (on desktop); Windows XP Professional SR 3 (on laptop) and Sennheiser MD 431 II mic with the Andrea USB pod on both computers; LAPTOP: HP Compaq 8710p; Intel(R) Core(TM)2 Duo CPU T7300 @ 2.00 GHz 2.00 GHz; 2.00 GB RAM; DESKTOP: Intel(R) Core(TM)2 Quad CPU @ 2.66 GHz 2.67 GHz; 2.00 GB RAM.

Dana,

According to Chuck Runquist the optimal range is 18-28.  However the critical feature is the "noise floor" of the microphone plus sound device.  The noise level (as seen in the Quality screen graph of the Audio Setup Wizard) should be as low and flat as possible with no spikes.  The speech signal should be high and again with a flat response across the frequency range.  To my mind this graph, together with the play button of the ASW, gives a very clear indication of the microphone/"soundcard" quality.  The low noise floor is more important than an overall high "quality" reading.

Graham 

 

-------------------------

 

www.itspeaking.co.uk

Apologies for double post.  It appears that when the page expires immediately after creating a reply a second copy of the post is created.

Graham

-------------------------

 

www.itspeaking.co.uk

Dana,

The most important line in Graham's excellent response is, "The low noise floor is more important than an overall high "quality" reading."

The next to last step with the blue bar adjusting the volume level is also important to watch. In this step it is not important where the bar starts out as Dragon is adjusting the audio level it requires for best recognition. The important thing to look at is the bar level when this step finishes. It should be less than 3 notches from the top or more than 3 notches from the bottom. 3 notches is about 20%. If the blue bar is close to the top the microphone likely has a weak output. Too close to the bottom and the signal from the microphone is too strong. When eMicrophones tests a microphone, this is one of the first things we look at to see if the microphone has a chance of success.

 I won't repeat Graham's excellent description of the last step with the graph as that was said so well. However, an old trick used by experienced Dragon users is to pause 3 or 4 seconds after clicking the Start Quality Check button in the last step of the Audio Setup. This gives the system time to sample the background noise and noise inherent in the system from things like the sound card and the internal computer components. When you do start speaking, the system has a good base of the background noise with which to compare against your voice level.

Try running the last step of the audio setup without a pause. Then run it with a 3-4 second pause before starting. I just tried it and my audio setup signal-to-noise ratio went from 23 to 26.  If you want an even higher number, make sure to enunciate every word as clearly as possible and give a good pause, at least one second, as you speak each phrase in this last step.

As you are aware, from using the Sennheiser MD431II, a good noise canceling microphone will be even slightly more accurate than a lesser noise canceling microphone even when there is complete silence. Of course there is even a more pronounced higher level of accuracy with a good noise canceling microphone when there is greater and greater background noise.

Marty

-------------------------

Martin Markoe - BANNED USER: This user has been banned from these forums

I would add a couple of quick points to Marty's explanation.

First, the quality check ratio (or Speech-to-noise ratio) is merely a check.  While the methods that Marty points out may give you a higher Speech-to-noise ratio, this stage in the audio setup wizard does not change any of the settings relative to your microphone.  Therefore, pausing 2 to 4 seconds before beginning speaking does allow the system to adjust to the background noise and will produce a better quality check ratio (speech-to-noise ratio).  However, this is only a measurement not an adjustment.  In other words, it gives you a better measurement of how well your sound system and microphone are able to distinguish between signal (speech) and noise (background), but it makes no changes to your microphone adjustment.  Whatever the volume adjustment made in the previous screen remains the same regardless of the quality check ratio that you obtain.  It is important to understand this because assuming that a better quality check ratio has improved your microphones efficiency is a false assumption.  The only way that you can change and/or improve your microphone settings is to go back to and rerun the "Start volume check."

Second, the optimal quality check ratio is generally between 22 and 28.  ANC microphones (Active Noise Canceling), such as the Andréa ANC 750, generally show a quality check ratio between 18 and 22.  This is normal and expected because of the way these types of microphones work.  Marty is right on the money with regard to good noise canceling microphones.  I get an average quality check ratio of between 26 and 34 using the Sennheiser MD431 II.  However, what is important to understand as regards the quality check ratio is that as the quality check ratio increases beyond 26, accuracy improvement plateaus.  That is, there is not that much of a significant gain in overall clarity of signal and accuracy between 26 and 34.  Still, if you're getting a very high quality check ratio that's good.  It means that your signal is very clear.  However, you will generally not see a significant improvement in accuracy with a quality check ratio higher than 28.  On the other side of the coin, if you were to graph the accuracy curve based on the quality check ratio, you would see a steady and significant rise in accuracy with a quality check ratio between 22 and 28.  You would see a very steep drop in accuracy between 22 and 16 (exception being ANC microphones).  And, you would see a virtually flat curve for accuracy gains between 28 and 34.

Lastly, the single most important characteristic is, however, the noise canceling capabilities of the microphone that you are using.  In other words, if you have a microphone such as the Sennheiser MD431 II, which has about the best noise canceling capability that there is available for speech recognition, then overall accuracy will be significantly better than a poor noise canceling microphone even if both microphones get exactly the same quality check ratio.  Accuracy with high quality noise canceling capability remains constant as the background noise increases, or at least relatively constant.  For example, in my testing with the Sennheiser MD431 II, the quality check ratio with a normal background noise level of 52 dB is between 28 and 34.  With the background noise level increased to 93 dB + (tradeshow floor background noise level) or 85 dB (CNN headline news playing through a Bose surround sound home theater stereo), the quality check ratio only drops to between 25 and 26, and the overall accuracy during dictation under these conditions remains fairly consistently the same as with normal noise level backgrounds of 52 dB.  On the other hand, if you perform the same test with the Andréa microphone that comes boxed with DNS, the accuracy is fairly good in a 52 dB background noise environment.  However, at 93 dB + this microphone fails miserably in terms of accuracy and the best quality check ratio you can get under these conditions is about 16.  Therefore, good, high-quality noise canceling capability is preferable to a high quality check ratio, but obviously having both is best.

Chuck Runquist
Former DNS SDK & Senior Technical Solutions PM for DNS

"We are all victims of mythology in one way or another. We are the inheritors, and many times the propagators, of a desire to believe what we want to believe, regardless of whether or not it is true." -- J.V. Stewart

 

-------------------------

Chuck - Thanks once again for your extremely insightful explanation of the quality check ratio! 

But - if I understand correctly - then why - if the noise level of the room remains consistent - day after day - do I sometimes get a 19 - and other times a 26 using the exact same set-up - and the exact same mic (Sennheiser MD431 II) in the exact same room???  That's the only thing that confuses me.  (Yes, Marty, I learned long ago to pause before starting to speak, but even when I do that - the results swing drastically.) 

Since my mic is on a boom - the only thing I can guess that changes these readings is the position of the mic.  And - what I do is to swing the mic over - below my mouth by about 2"-3" and slightly to the side and then I talk directly down on the top of the mic while it is is a true vertical position.

I also learned long ago not to pay any attention to these readings at all.  Since I consistently obtain 99% - 100% accuracy almost all the time, I guess I can continue to ignore these readings, right?  :-)

 

-------------------------

Dana Joan - Vero Beach, FL  - 

Dragon NaturallySpeaking Medical, Version 10.1; General Medical Large Vocabulary; Windows 7 Ultimate (on desktop); Windows XP Professional SR 3 (on laptop) and Sennheiser MD 431 II mic with the Andrea USB pod on both computers; LAPTOP: HP Compaq 8710p; Intel(R) Core(TM)2 Duo CPU T7300 @ 2.00 GHz 2.00 GHz; 2.00 GB RAM; DESKTOP: Intel(R) Core(TM)2 Quad CPU @ 2.66 GHz 2.67 GHz; 2.00 GB RAM.

Chuck's extremely insightful explanation has been included as a reference in the Speechwiki article SpeechToNoiseRatio. I try to immediately integrate them into the my articles when I see them.

Ian Nastajus

-------------------------

Former primary author of Speechwiki NaturallySpeaking FAQ.