Vowel Sound Imagery Decoding by a Capsule Network for the Design of an Automatic Mind-Driven Type-Writer

This paper intends to develop a novel methodology for modeling a mind-controlled type-writer system to cater the needs of individuals suffering from various communication related disorders. This objective is fulfilled by first capturing EEG signals from ten subjects involved in mental utterance of seven vowel sounds. The eLORETA analysis of the acquired signals confirms the involvement of occipital, parietal and prefrontal lobes for this cognitive activity. The procured signals undergo the process of feature extraction after eradication of artifacts and are transferred to a novel capsule network module for categorization of seven class labels. Performance analysis undertaken confirms the superlative behavior of the proposed classifier with respect to other standard ones. Moreover, statistical evaluation also assures the superior performance of the proposed classifier model. A coding scheme has also been proposed to signify the consonants by the coalescence of two vowel sounds segregated by a space. Thus, the proposed methodology can be effectively utilized as a mind controlled type writing system to serve the needs of disabled individuals. Additionally, this technique can also be used in certain military scenarios that demand non-verbal communication as a secure option and in various BCI based gaming applications.