Acoustic Optimization in Dragon
If you’ve not run into it before, the Acoustic and Language Model Optimizer, hereafter referred to as the ALO, is a module that allows corrections, training, and dictation to be incorporated into your medical profile. By regularly (and correctly) running this process in Dragon Medical Practice Edition 2, you can boost your recognition accuracy to greater and greater levels—even up to 99%. This, of course, frees you up to fill out patient charts quickly, dictating medical notes with confidence.
So, then, let’s take a look at how the process of accuracy tuning works, as well as touch on a few key issues with regard to utilizing the ALO function properly. Please note that we’ll only be covering procedures relevant to Dragon Medical Practice Edition 2 users running local profiles. We will cover roaming optimization in another post.
The Basics of Dragon Accuracy Tuning
In order for the process of optimization to work properly, there are a few fundamentals that you need to observe. For instance, if you were to just guess at how the ALO tool works, what would you come up with? That it needs to hear enough of your voice before it can improve its accuracy? That it needs your help to determine whether it’s giving you the correct results? Perhaps, even, that it would like to know how you use words to construct phrases?
You’d be correct in all three cases. The ALO, in order for it to truly improve your recognition accuracy, requires that you dictate generously and make corrections as needed. You may ask why, as I did.
In order for DMPE 2 to change its impression of your voice, it needs something to compare with. Framed slightly differently, if you dictated, “Spina bifida is a major concern with this patient,” but Dragon produced something else, how would it know it was wrong? Dragon must be corrected in order to fix this mistake. And if there’s a particular word that you need Dragon to get, especially something unique, like a doctor’s name, that word must introduced into the vocabulary and trained. Then the software knows to anticipate the word, and can listen for your pronunciation of it.
Additionally, Dragon Medical Practice Edition 2 gets a lot of useful data when you read the training text, available by going to Audio > Read text to improve accuracy from the DragonBar. This training can be especially powerful, because the stories Dragon asks you to read are known quantities; whereas your voice, including its unique inflections and intonations, is an unknown quantity. The more you read, the more the unknown quantity is mapped to the known. Think of this process as similar to training individual vocabulary words—but on a much larger scale.
While you’re supplying DMPE 2 with all of this data, dictating and correcting, it’s storing away audio files for the time when optimization is run. So, if you use Dragon regularly, it’s recommended that you run this process once a week. That way, improvements can be made to your acoustic and language models—both of which are necessary for accurate speech recognition.
Click here for more information on how Dragon Medical Practice Edition 2 learns your voice.
How to Schedule Dragon Profile Optimization
There are two ways to go about kicking off the ALO task. But before we get into it, let’s discuss how to set it up and what to expect.
The process of optimization can take quite a long time—depending on how much data there is available and how powerful your workstation is. You can expect to not be able to use your computer for between thirty minutes and two hours, as accuracy tuning is processor intensive. This is why we usually recommend running it in the middle of the night.
When optimizing, you should select both the acoustic and language models. This ensures the most improvement to your recognition accuracy. You’ll also need to make sure your computer stays on and is not set to go to sleep or hibernate. Otherwise, you’ll get the dreaded “Dragon couldn’t run tasks to improve speech recognition” error.
Scenario 1: Run Immediately
If you’re looking to start optimizing right away, look at the right-hand side of the DragonBar. Click on Audio > Launch Accuracy Tuning. That’s pretty much it, with a small exception.
Open Tools > Administrative Settings from the DragonBar and click “Proceed.” Click on the “Scheduled Tasks” tab and make sure that the “Let the user choose” box is checked at the bottom of the panel.
Scenario 2: Scheduled
This method requires a little more configuration…and TLC. While accomplishing the same thing, there are a few extra steps to take here.
Now, Dragon Medical Practice Edition 2 does give you the chance to set a schedule for accuracy tuning during profile creation. But at that time, you’re probably still getting your bearings, and are likely to have glossed over that part. No worries.
No matter what, you’re going to need to check four boxes for scheduled optimization. First, go to Tools > Administrative Settings on the DragonBar, and click the “Proceed” button. Open the “Scheduled Tasks” tab and check the “Enable scheduled Accuracy Tuning” box.
Next, click on the “Configure” button and make sure that both tabs reflect your scheduling choices. You should probably space out the two by about an hour. Also ensure that the “Enable” boxes are checked for both the acoustic and language model.
Finally, open your profile and navigate to Tools > Options > Data. Make sure that the “Run accuracy tuning at the time scheduled by your Administrator” box is checked. Please note that Dragon must be closed at the time you scheduled for optimization.
ALO Best Practices
There are a few things you need to do to make sure that the accuracy tuning process goes off without a hitch. These tips and best practices come straight from the folks at Nuance.
- On the DragonBar, navigate to Options > Data, find the checkbox next to “Conserve disk space required,” and make sure it is not checked. This allows Dragon to save your audio files, which are necessary for acoustic optimization.
- In the same panel, check the “Store corrections in archive” box. This ensures that DMPE 2 incorporates your corrections in the optimization process.
- Finally, Options > Data gives you the option to change the amount of audio data Dragon stores. Click on the “Archive size” button and make adjustments as needed.
- Although you can interrupt optimization, it’s best to just let it finish. And don’t get your hopes up by looking at the progress bar: it’s unreliable as a measure of time remaining for the ALO task.
- As we mentioned earlier you must perform corrections or do some training in order for optimization to work. Additionally, any corrections you make need to be trained. Otherwise, Dragon will tell you don’t have enough data to optimize.
- Before starting, the optimizer backs up your profile. If you cancel out of the accuracy tuning process before it’s done, DMPE 2 reverts to this backup automatically.
Addressing Roaming Profile Optimization
We covered a lot of ground here today—but, believe it or not, there is more. Optimizing a roaming profile requires another set of instructions and procedures. As mentioned, we’re saving that for another day.
Edit: That day has come. Click here for instructions on scheduling roaming profile optimization, or here to find out more.
In the meantime, if you’d like some one-on-one virtual assistance with this, or any other procedure within Dragon Medical Practice Edition 2, why not sign up for our technical services. We’d love to assist you with making speech recognition easier, whether it’s related to workflow, troubleshooting, or accuracy.