The project described in this proposal involves the design and implementation of a speech driven computer interface which could be used as a communication tool for the disabled. The approach outlined involves the use os parallel filter banks, followed by various signal processing stages and a gamma network composed of gamma memory sections and a back-propagation neural network. The system will be designed for use on a NeXT computer, which contains a microphone and a DSP chip, so that no additional hardware will be required. The investigators will test the system on isolated words although one of the justifications for the neural network approach is so that a continuous signal stream could be processed without necessitating the segmentation of the speech signal into allophones or words. It is expected that this approach will perform better than techniques that are presently used on low cost computers. Because this approach uses networks designed to process time-varying signals in a speech recognition system it could have an impact on the technology in this area if it can be extended to process continuous speech.
