Exploring the topographical and time-frequency representations of electroencephalography (EEG) for inner speech decoding

Inner speech is an internal language that expresses thoughts and ideas within the mind, without producing sound. Recording inner speech activities through scalp electroencephalography (EEG) provides insights into the mechanism of language production. Our current study aims to explore the neural mechanisms of inner speech among Chinese native speakers through collecting EEG signals during inner speech tasks, and extract topographical and time-frequency related features for decoding the brain's language formation. Specifically, we designed a picture-induced speech and non-speech task. EEG signals were collected from 7 subjects across three different tasks: inner speech, pronounced speech, and visual tasks. Through computational analysis of event-related potential and time-frequency domain features, comparisons were made regarding power discrepancies among different tasks and words across six frequency bands. Power difference topographic maps were also generated to compare spatial disparities between tasks and words. Our results indicate that the inner speech in Chinese Mandarin shares a similar articulatory encoding process with pronounced speech. After the first second of inner speech execution, the spatial distribution of gamma power may help distinguish directional words such as "up", "down", "left", and "right". This research provides insights for further exploration of inner speech decoding among Chinese native speakers.