What I did was opened Dragonpad and read a passage of about 3 sentences as slowly as possible (almost stopping in between words), then I read it slightly faster, then slightly faster than that, then finally at my regular pace. This improved my accuracy by exactly 10 percentage points. I only had to read the passage about 4 or 5 times.

DNS 9.5 Pro, KnowBrainer 2007, Intel Quad CPU 2.4Ghz, 3MB RAM, 300GB HD, BW900 mic. Email: greatgrains@hotmail.com

Univoice is now 2% more accurate for me than my current user! I worked with this for a while speaking very slowly and then up to normally.

When I first tried UniVoice it got 91.6% accuracy. Now when I speak normally it is 2% more accurate than my current user.

I tested the accuracy several different times, and finally tested it with an independent document of about 600 words. Unvoice got  97.9% accuracy, and my regular user got 95.9% accuracy.

One other thing that I think is out of the ordinary is that I know some people were having problems dictating the word 'UniVoice.'  When I say 'UniVoice' it spells perfectly for me.  I did notice that it is in my custom words dictionary, but I didn't do anything to put in there.  It was just automatically there.

DNS 9.5 Pro, KnowBrainer 2007, Intel Quad CPU 2.4Ghz, 3MB RAM, 300GB HD, BW900 mic. Email: greatgrains@hotmail.com

All you are testing here is DNS.  Regardless of the user profile that you're using, speaking at a normal pace is what produces better accuracy.  It has nothing whatsoever to do with whether you're using UniVoice or another profile created using DNS.  I use the untrained user profile (I have never train it for two years using the general training) and I get 99.7% or better virtually all the time.  Then again, my normal pace of speaking is about 160 words a minute.  In short, is how effectively you use DNS (i.e., whether you use it as it was intended to be used or you abuse it) that determines the accuracy.

UniVoice is not any more accurate than the standard DNS user profile as long as you dictate and correct appropriately and as long as you're hardware is optimal.  I've had the current UniVoice 9.5 since Lotus released it and I spend as much time dictating in that user as I do in my standard DNS user and my DNS user is approximately 1.5% more accurate than the UniVoice 9.5 user.

My point is that it is your Acoustic Model and the way DNS works that determines the accuracy level that you get any user profile regardless of its source.  Unfortunately, your test doesn't prove anything except that you have discovered how DNS works best.

Chuck Runquist
Former Dragon NaturallySpeaking 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

I may have misundertood you, but I have created more than one user profile in DNS in the same way that the Univoice profile was created (a blank slate profile without any training), and I didn't get near the accuracy that I got with the Univoice profile.

In fact, the user profile that I was using (before I changed to the Univoice profile, which has 2% better accuracy) was created as a blank slate, without any training.

DNS 9.5 Pro, KnowBrainer 2007, Intel Quad CPU 2.4Ghz, 3MB RAM, 300GB HD, BW900 mic. Email: greatgrains@hotmail.com

All that indicates is that UniVoice works better for you given your dictation style, volume of your voice, timbre, pitch, etc.

Because of the way that Lunis created UniVoice, it will generally start out more accurate than a new user created using the DNS profile with the "Skip initial training for this user" for some users.  For others, the opposite is true.

However, in my testing over the last 3 years, the advantage of UniVoice doesn't last very long if both profiles are used equally over time.  UniVoice caters to users whose dictation style, volume, timbre, pitch, etc. are more characteristic of non-professional speakers.

Nevertheless, and I'm not trying to discourage anyone from using UniVoice, what makes any user profile more or less accurate is the quality of the Acoustic Model.  UniVoice starts out with a little broader support for a wider range of speaker types than DNS.  The corpus of speech data used to create the DNS independent speaker profile (i.e., no training) uses a broader base of professional speakers (some 10,000 plus contributors with over 10 million words accompanying such), but balances gender.  I don't think Lunis has access to this much speech data.  On the other hand, DNS does not generally include as broad a base of the younger speakers and older speakers.

Still, let's not make this issue a shootout at the OK corral.  Suffice it to say that if the UniVoice works better for you then the adage applies "if it ain't broke, don't fix it."  On the other hand, don't make the claim that a UniVoice profile is better than the DNS profile because the results that you get using UniVoice are unique to you.  There are posters on this forum that are equally adamant that UniVoice stinks.  UniVoice works well for some and not so well for others.  On the other hand, the DNS independent speaker profile works better for some and not so well for others.  All I'm saying is it doesn't follow that claims to one profile being better than the other.  Claims to this effect are unsupportable on both sides of the fence because the main variable is the user not the profile.

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

"It ain't what you don't know that gets you into trouble. It's what you know for sure that just ain't so." -- Mark Twain

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