12.0

Chapter 12: Improving recognition accuracy
optimization.
n Be sure that the "Store corrections in archive" option is selected on the
Data tab of the Options dialog box. Otherwise you may see a message
that there is not enough data for the Acoustic and Language Model Opti-
mizer to process. The "Store corrections in archive" option should be
selected whenever you use Dragon for an extended period of time.
n If you have a User Profile set to the Correction Only Mode (for example,
a transcriptionist), you must disable the Correction Only Mode option on
the Miscellaneous Tab of the Administrative Settings dialog box before
running the Acoustic and Language Model Optimize on this User Profile.
To run the Acoustic and Language Model Optimizer
1. Say "Switch to DragonBar," then "Audio," then "Launch Accuracy Tuning,"
or click Audio> Launch Accuracy Tuning on the DragonBar.
The Acoustic and Language Model Optimizer dialog box opens with Perform
Acoustic Optimization and Perform Language Model Optimization selected.
Nuance recommends that you run both optimizations at the same time to
maximize accuracy.
2. Click or say "Go" to start the Acoustic and Language Model Optimizer.
3. The Acoustic and Language Model Optimizer may ask if you want to save
your current User Profile. Click or say "Yes" if you made corrections since
the last time you ran the Acoustic and Language Model Optimizer, or if you
have not run the Acoustic and Language Model Optimizer since you started
using Dragon.
The Acoustic and Language Model Optimizer dialog box shows Dragon's
estimate of how long it will take to finish adapting your User Profile. How long
this takes depends on the amount of Acoustic Data stored in your User Profile.
Notes
n By default, the amount of acoustic data saved is limited to 180 minutes
(the maximum). You can reduce this amount on the Data tab of the
Options dialog box.
n If you attempt to run the Acoustic and Language Model Optimizer before
you have trained any corrections (even if you have corrected errors with-
out training the corrections), or before you have performed any addi-
tional training, you will see an error message saying that there is not
enough data for the Acoustic and Language Model Optimizer to process.
n Training multiple words or phrases rather than training individual words
is generally more effective at increasing accuracy. Most people have a
tendency to over enunciate when saying individual words as opposed to
the more normal speech patterns of saying a complete phrase.
n The Acoustic and Language Model Optimizer automatically makes a
backup of the old User Profile before it begins modifying them. Your User
204