Word Watch: An Open-Source Application for Creating Read-Word Data for Subvocal Recognition Tasks

Abstract

Subvocal Recognition (SVR) has the potential to revolutionize communication for those with speech impairments and enhance human-technology interactions. However, current data collection methods for SVR are time-consuming, laborious, and difficult to scale. This paper presents Word Watch, an open-source application that offers a novel approach to SVR data collection by targeting subvocalizations during reading. Word Watch employs a commercially available eye-tracker, calibrated text-settings, and custom algorithms to predict which words a user has read and when those words were subvocalized. Evaluation of the Word Watch system shows that it can accurately identify which words have been read at a reasonable reading pace. Further research, however, is needed to validate the application's estimates of when read-words are subvocalized.