User Manual

102
Creating Operator Voice
Creating Operator Voice Creating Operator Voice
Creating Operator Voice
Te
TeTe
Templates (enrollment training)
mplates (enrollment training)mplates (enrollment training)
mplates (enrollment training)
• Purpose
• Before You Begin
• Procedures
Purpose
In order for the Talkman® terminal to understand what an operator says, the operator must
create a voice template through a process called enrollment training.
Before You Begin
During enrollment training, the Talkman terminal gives the operator words to say, and the
operator says the words to the Talkman terminal. The terminal prompts the operator for each
word in the task at least four times. The software then stores one version of each word in
template files for that operator.
Note: When an operator is training a digit, the operator will not only have to say the digit
four times, but he or she will also have to say all of the embedded digit strings that contain
the particular digit. Embedded digit strings are strings of numbers that contain the digit
that is being trained. For example, for the digit "2", the embedded strings might be 020,
212, and 222. When training the embedded digit strings, the digits should be spoken
without pauses between them. The operator should speak the digits as naturally as he or
she would when telling someone a phone number.
Also, if the task that is being run contains decimal points (i.e. the word "point"), the
embedded digit strings will contain strings with the word "point" in them (2 point 2).
An operator can either hear or read the words to say during enrollment training. In order to
hear them, operators simply listen to the words through their headsets. To train by reading the
words, operators use either a visual training device or a computer running Display Manager. For
details about Display Manager, refer to the Display Manager topic found in the Talkman
Management Software online documentation. Vocollect recommends training operators using
the visual training device, as operators are more likely to speak in their natural voices when
they read the words than when they hear them. Operators should try to speak as naturally as
possible while creating their templates.
Allow adequate time for operators to complete this process. The amount of time required will
depend upon the length of your task; allow about 40 minutes for every 100 words in the task.
Once you have trained a few operators, you will have a good idea of the amount of time to
allow for future enrollment training sessions. It is best to create the voice templates all at
once; if an operator stops training, the Talkman terminal will time out and turn off (just as it
does during regular use). If this happens, pressing the yellow play/pause button will restart the
terminal.
For best recognition results, the vocabulary words should be trained in the same noise
environment in which the terminal is going to be used. However, this rule only holds true if the
environment does not have a lot of loud non-stationary sounds (i.e. pallet drops, horn beeps,
PA system, etc). If the environment in which the terminal is going to be used does have a lot of
these types of noises, then better results will be achieved by training the unit in an
environment where these noises are not quite as prevalent.
Prepare the terminal before the operator begins. Load the task (see Loading a Task Onto a