Glove-talk: An Adaptive Interface That Uses Neural Networks
Sidney Fels · 2005
An interface which converts hand gestures to speech (Glove-Talk) was created to investigate the use of neural networks for adaptive interfaces. Glove-Talk is implemented with 5 neural networks and has a vocabulary of 203 words with variable speaking rate and stress. Glove-Talk is trained for a single user using automatic learning procedures It says the wrong word less than 1% of the time and says no words about 7% of the time. Current work is underway which looks at a finer grained mapping of hand gestures to speech and thus requires a more complex interface.