On Thu, 7 Mar 2002 15:41:46 -0500, Penny Gray <[log in to unmask]> wrote:

>I have recently begun working with a deaf student who has a repetitive
>stress injury which precludes her use of a keyboard, in a postsecondary
>institution, in the use of Dragon NaturallySpeaking.  She has been deaf
from
>birth but has very good and understandable (by human ears) speech.  We have
>gone through two trainings in which Dragon did have trouble understanding
>some of her speech, even to get through the training sessions.  I think
that
>her speech is good enough that Dragon should be able to understand what
>she's saying if it is trained enough.  I was wondering if anyone has had
any
>similar experiences and possibly some suggestions.  I don't want her to get
>too frustrated in the efforts to train Dragon.

Some suggestions:

- Make sure the PC sound system is good enough to run speech recognition
software. One way to do this is to create a user for someone WITHOUT a
hearing impairment. Run Audio Setup Wizard and check the Signal to Noise
ratio. If the value is low (say, 16 or 18), then the PC sound system is
probably inadequate to run speech recognition for anyone, let alone someone
who is deaf. (Hopefully, you will obtain a S/N value of over 20.) If the
sound system proves to be OK, you can delete the test user. Otherwise, upgrade the sound card and/or replace the microphone. Note: identical
microphones can have vastly different sonic characteristics. I recently
ordered five identical VXI microphones. Two hardly worked, two were less
than adequate, and one was the most accurate microphone I have ever tested.
There are software tricks for boosting S/N, but I have not found them
especially useful.

- Avoid using NatSpeak Version 6 unless the PC has an extremely fast CPU
and oodles of RAM. 256 MB RAM is a bare minimum. For version 6, I would
double the RAM. Version 6 offers some enhancements over version 5, but I
have found enough problems with Version 6 that I would not recommend it
until the manufacturer fixes the more glaring bugs.

- Training may be difficult because the Audio Setup Wizard times-out if a
user is silent or does not make sounds that the system is expecting. The
time-out is NOT adjustable, as far as I can tell.  (Grrr!!!) You can help
the student get through training by giving her opportunities to practice
reading the texts. Do this: Find the file that contains the training
materials that you need. For example, if you decide to use Kennedy's
inaugural speech, it is contained in ...NatSpeak\Training\enx\data34.bin.
Make a copy of the file, rename it “data34.txt” or “data34.doc,” open the
renamed file in a word processor, format it nicely, and print it out. The
student can practice reading the texts, and at the same time, you can coach
the student on proper dictation techniques. (I am training someone with a
visual impairment to use NatSpeak. I had her train the system by reading a
copy the training text printed in 24-point Arial bold rather than from the
screen.) You may also need to print out the texts from the two Audio Setup
Wizard tests. I have appended these texts to the end of this message.

- When you finish training, make sure you run the Vocabulary Builder on two
types of files:

1. A list of expressions that the student uses, e.g.,

Joan Q. Public
Developmental Psychology
Canada Customs and Revenue Agency
Journal of Medieval Literature
[log in to unmask]

2. Several samples of the student's writing: email messages, essays,
letters, whatever.

Running the Vocabulary Building in these ways significantly increases
accuracy.

- I could add more about training someone to use NatSpeak with minimal use
of the hands, but that is for another message. See if you can get through
the training first!

Alan Cantor
Cantor + Associates Inc.
Workplace Accommodation Consultants
[log in to unmask]
www.interlog.com/~acantor


Volume test:

In this step the computer listens to the sound of your voice and adjusts
the volume setting of your microphone. When the computer has finished
adjusting the volume, it beeps to signal that the process is complete. If
you reach the end of this text but you have not heard a beep, start reading
the text again from the beginning. You should only have to read for about
ten to fifteen seconds.

Sound quality test:

In this step the computer checks the audio input from your sound system.
Having high-quality audio input is very important for good speech
recognition. Poor audio input will make it difficult or impossible for the
program to recognize your speech accurately. When the computer has finished
checking the audio quality, it beeps to signal that the test is complete.
If you reach the end of this text but you have not heard a beep, start
reading the text again from the beginning. You should only have to read for
about fifteen seconds.