VR Class Projects: Speech Recognition
=====================================

Install and play with the speech recognition software. Try to answer
the following questions:

+ Can you train the software to understand additional words? Is the
software speaker dependent or speaker independent? Can you store
different pronounciations of one word into the same database so
a spoken word is compared with different pronounciations?

+ How good is the performance, i.e, the percentage of correctly
identified word, before training - 70%, 90%, or > 99%?
And after training? Does this depend on the speaker/accent?

+ What does an additional new speaker have to do to be understood?
How long does the required training take in average? Can we avoid
additional training? How good is the performance if no training
is done at all but using your extended database. Do some tests
and ask other students and faculty to volunteer.

+ Which words to you want to train to the software? Prepare a list.
Work together with the Mini-Cave group. Then train these
words/commands to the speech recognition software. Also consider the effect
of two possible listening modes: one, where every word you say
is interpreted, and one where only words following the keyword
"Computer" are interpreted.

+ What is your personal experience with the software? What do you
think of its speed? Do you like it? Would you recommend its
future use? Are there different opinions within your group? -
If so, let us know about all these opinions!

+ Now the technical part of this project: Where does the software
store the recognized words? Can you redirect this output?

+ Develop some software that makes use of interprocess communication (IPC)
technology that passes the identified word from one computer system
to another. This should work both in a homogeneous and in a 
heterogeneous environment, e.g., from PC to PC or
from PC to SGI. A client-server approach based on sockets or
the use of Remote Procedure Calls (RPCs) might be best. Let the
other system output the received words.

+ BONUS: If time permits, work with the Mini-Cave group to combine
both projects, i.e., have the speech recognition software control
the Mini-Cave application.

Turn in a typewritten report (10-20 pages of text, plus appendix
such as computer code, computer output, protocols of experiments)
by MONDAY 4/27/98. Report everything that might be of interest.
If you planned to do something but couldn't realize it due to
unexpected problems, let us know about these problems.

Prepare a 30 minute talk for presentation in class on 
THURSDAY 4/30/98. Also be ready to give a demonstration of
your work some time the same week.