CRG-TR-90-1 request

[Carol Plathan]

         PLEASE DO NOT FORWARD TO OTHER NEWSGROUPS OF MAILING LISTS
	 **********************************************************

The following technical report is now available.  If you'd like a copy please
send me your real mail address (omitting all other information from your
message).

---------------------------------------------------------------------------

             BUILDING ADAPTIVE INTERFACES WITH NEURAL NETWORKS:
		       THE GLOVE-TALK PILOT STUDY

                             S. Sidney Fels

                       Department of Computer Science
	                   University of Toronto
                          Toronto, Canada M5S 1A4

	                       CRG-TR-90-1

Connectionist learning procedures can be used to interpret incoming data and
to generate complex responses.  To illustrate the potential of using these
procedures for adaptive interfaces, a system using neural networks to convert
hand gestures to speech in real-time was developed.  A VPL DataGlove connected
to five networks and a DECtalk (speech synthesizer), implements the hand to
speech system.  Using existing connectionist learning procedures, the complex
mapping of hand movements to speech particular to a specific user is learned
using data obtained in a simple training phase.  Based on a 203
gesture-to-word vocabulary, the noticeable word error rate is less than 1%.
In addition, adaptive control of the speaking rate and word stress is
available.  The system is streamlined by using small, separate networks for
each naturally defined subtask.  Smaller networks reduce training and running
times.  This system demonstrates that connectionist learning procedures can be
used to develop the complex mappings required in an adaptive interface.

---------------------------------------------------------------------------