Why is it advantageous to have the "select" command bring up the Correction Box and to have the "correct" command bring up the spell box?

>>

Select-&-Say correction is about 3 times faster than Spell That correction and if you don't see the right choice in the window, you have the option of re-dictating over the highlighted text.

I thought that we were always required to use the correction box to correct mistakes, or else we would trash out our user files. If there are words which are wrong and I use Select-and-Say to select them and the correct words are not listed in the list of choices and I just dictate over them the same thing is that okay? Is my user filecorrected that way?  clear as mud?

---

Tuck Fexas

I don't know where you got the above information, but it's definitively incorrect.

Although there are some significant changes in DNS 9/9.1/9.5, to many users are under the misconception that DNS learns misrecognitions if they are not corrected. This is simply not true. The logic of the way that DNS works is that if you do not make a correction, DNS simply assumes that what you dictated was correct. This does not affect your user profile or corrupt it in any way because no changes were made to your user file. The only thing that occurs from failing to correct misrecognitions is that DNS will simply continue to make the same mistakes.

If you use the correction window or the Spell dialog to correct misrecognitions, DNS learns to associate your pronunciation(s) (links them) to the way that you pronounce them when you are dictating. If you train a correction, and you shouldn't train a correction if the original playback of your dictation is clear and correctly enunciated, then DNS will make appropriate changes to your Acoustic Model, along with linking that training to the appropriate correction, as well as making an entry in the Acoustic and Language Model Optimizer archive for your Language Model. However, the Language Model does not get updated fully until you run the Acoustic and Language Model Optimizer. However, changes are recorded to your user as appropriate when you save your user on closing a session, but saving your user does not update your Language Model(s) as that is done only through the Acoustic and Language Model Optimizer, or analyzing your sent e-mails or documents for adapting to your writing style. Note that there are several additional issues relative to how and when to train corrections. There are any number of posts on this subject in this and other forums. I'm not going to go back into great detail on these at this point.

As regards the other part of your question, correcting misrecognitions using Select-and-Say (i.e., select and re-dictate), DNS also learns from this approach by updating your Acoustic Model. In DNS 9 your Acoustic Model is constantly updated using the new SilentAdapt feature, which is equivalent to and basically the same as general training on a continuous on-the-fly basis. Therefore if you correct in this manner, you are essentially training your Acoustic Model to correctly associate your pronunciations with the corrected phrase. However, the way that this works is that if you select a word or phrase and re-dictate it, and DNS gets it correct, it means that DNS didn't clearly understand you (i.e. your pronunciation/enunciation) correctly the first time. This problem often occurs from dictating in short choppy phrases (i.e., 2 or 3 words at a time). If you dictate in short choppy phrases, your accuracy and the number of misrecognitions will increase. You should learn to dictate in full phrases of at least 5 to 10 words, and preferably longer. The more you dictate without pausing, the more accurate Dragon NaturallySpeaking will be, and the fewer misrecognitions will result.

It is critically important to remember one thing with regard to DNS and misrecognitions (accuracy). The clarity of your diction (i.e., avoiding mumbling, slurring your words, or running your words together), the quality of your hardware, and the more clearly you enunciate/pronounce words and phrases, the more accurate your dictation will be. In short, before you go blaming misrecognitions on the software, take a good look at your hardware and your own dictation style. In other words, people who live in glass houses shouldn't throw stones.

Chuck Runquist

Former Dragon NaturallySpeaking SDK & Senior Technical Solutions PM for DNS

"If problems can be created by knowledge, then they cannot be solved by ignorance." Isaac Asimov

---

Core2™ Extreme QX9650,  5 Core2™ Duo’s 2.0 GHz-3.06 GHz, Core™ i7 980X Extreme, DDR2 RAM 4 GB 1066 MHz Core2™ Duo systems, DDR3 12 GB 2000 MHz Core™ i7, VelociRaptor 10,000 RPM boot drives all systems.

First, I discussed running the Acoustic and Language Model Optimizer with UniVoice with KnowBrainer (i.e., Lunis). I informed him that I experienced no problems in doing so. The way that the Acoustic and Language Model Optimizer works, it should not have a negative impact on UniVoice.

Second, in DNS 9 (i.e., 9.0/9.1/9.5) the acoustic optimizer is enhanced to include both Acoustic and Language Model optimization. There are also a number of additional enhancements, primary among which is the fact that the Acoustic and Language Model Optimizer archive file now retains a complete and updated (i.e., every time you run the Acoustic and Language Model Optimizer) record of all your changes and adaptations. When the Acoustic and Language Model Optimizer is run it performs 3 functions:

1. It creates what is called a Delta file (i.e., Delta meaning change) which is a collection of the differences between the complete Acoustic and Language Model Optimizer and everything you have done as far as corrections, training, etc.

2. This Delta file is used to extract redundant and older changes and references and consolidate both your Acoustic Model and your Language Model(s).

3. It then adapt your user files with the revised Acoustic Model and Language Model(s), as well is prompting you to save your users to ensure the retention of this information in your user profile. Part of this process is also to update the complete Acoustic and Language Model Optimizer archive to reflect these changes.

The end result is that the DNS 9 Acoustic and Language Model Optimizer is more comprehensive and efficient. It definitely will improve the accuracy of your user. However, keep in mind that the degree of improvement is relative. If you're getting 99.5% accuracy, the Acoustic and Language Model Optimizer may only improve your accuracy by .05 to .1% +/-.025% each time you run it. But it will improve it. Nevertheless, you will arrive at a point where the improvements provided by the Acoustic and Language Model Optimizer may reflect only certain changes. The reason for this is that SilentAdapt is constantly modifying your Acoustic Model. Therefore, improvements resulting from running the Acoustic and Language Model Optimizer may only reflect and impact on specific words and phrases, or specific Language Model modifications. The basic purpose of the Acoustic and Language Model Optimizer is to create an overall archive that accurately represents your optimal user profile. In the long run, this helps to prevent your user files from suffering accuracy degradation. Therefore, this entire process involves both optimizing your user files and maintaining their optimization. It also acts like a defragger on your hard drive. That is, it removes redundant and invalid information, as well is consolidating your user files. So the term "Optimizer" is really the key word in the entire process.

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

Education is what survives when what has been learned has been forgotten. - B.F. Skinner (1904-1990)

---

Core2™ Extreme QX9650,  5 Core2™ Duo’s 2.0 GHz-3.06 GHz, Core™ i7 980X Extreme, DDR2 RAM 4 GB 1066 MHz Core2™ Duo systems, DDR3 12 GB 2000 MHz Core™ i7, VelociRaptor 10,000 RPM boot drives all systems.

hello again, Chucker,

I ran the acoustic optimizer last night.  Now, UniVoice does not recognize my voice.  It recognizes no commands and when I dictate into it it is gobbledygook.  I am dictating on Dragon 9.5 and it is very accurate.

any ideas?

as always, your input is much appreciated. 

Jim 

 

 

 

 

---

Triona

thank you for getting back to me, Chucker,

I am using UniVoice 9.1.  I have rebooted several times just to be sure.  Still acts as if I am dictating in Vulcan. 

we can probably now expect Lunis to jump in and tell us "I told you so!"

should I reload UniVoice?

much appreciated.

Jim 

 

 

 

 

 

 

---

Triona