An automatic input protocol recommendation method for tailored switch-to-speech communication aid systems

Abstract

A switch-to-speech interface can provide a means of interactive communication as a support system for people with disabilities with voluntary movements. Any motion of a part of the body, such as eye movements, can be used for the switch input. The number of possible switch operations varies from person to person, but the bandwidth is generally quite limited. Therefore, efficient input protocols are needed to map the switch operations to pronunciations. Meanwhile, the protocol must be easily learnable so that anyone can use it. To this end, we propose a protocol recommendation method that can accept individual requirements in switch operations. This method suggests a customized protocol for each user of the interface that is both speedy to enter and easy to remember. The two main ideas in the protocol design are utilizing the knowledge about the alphabet table that everyone already knows and improving the input speed and learnability by allowing ambiguity in the switch to pronunciation conversion. The conversion errors due to the ambiguity are offset by an N-gram language model. The performance of the protocols was evaluated through simulations and the measured values obtained from research participants, and the advantage of the proposed method is shown.