A hybrid neural network based on the small-world network for inner speech recognition

Abstract

This study is devoted to the identification of human inner speech using an electroencephalogram (EEG), where inner speech refers to an individual's subjective experience of language, disconnected from discernible audible articulation. The core aim of this system is the rapid and precise classification of signals via human inner speech, thus facilitating enhanced control and interaction functionalities. The research entails a comprehensive analysis of 10 volunteers' brain activity across 128 channels from OpenNeuro's Inner speech dataset. A hybrid neural network, which incorporates the small-world network structure, is employed to model neural activity within the brain. This approach outperforms random chance and aligns with current research expectations.