Studying human factors for dwell-free text-to-speech communication systems

Abstract

People with severe disabilities that have difficulties communicating may find it easier to do so using a text-to-speech application with an eye-tracker. In this research, human factors related to dwell-free eye-typing techniques for on-screen keyboards are studied. A multi-phased study is conducted in order to develop an efficient text-to-speech application. During the first stage, performance of an existing dwell-free eye-typing technology, as well as associated keyboard layouts: the QWERTY, Alphabetical and DF-AJR will be evaluated. The Longest Common State Mapping (LCSMapping) algorithm is one of the best string processing algorithm, however it has its limitations. Proposals are made to improve accuracy, for which a specific research question has been devised. An N-gram language model is used to analyse which word is most likely to occur in the context given by the user. This study shows that there are slight improvements in mean accuracy when introducing the N-gram language model with the LCSMapping in all the keyboard layouts however the statistical tests deduct that the difference is not significant. In the second stage of this study a text-to-speech application incorporating dwell-free eye-typing technology is developed following a user-centred design methodology. The development of the tool was done over five iterations with domain experts and end-users. Feedback from each iteration was taken into consideration and modifications to the application were done to achieve a TTS keyboard application which enhances the user experience with the benefits offered by the dwell-free typing technique.