Spatial, Spatiotemporal, and Feature Analysis of Electroencephalogram During Imagined Speech with Visual vs Audio Cues

People with speech disorders have difficulties in social life and welfare. To overcome this barrier, researchers have conducted studies to recognize imagined speech from biosignals. The essential biosignals in producing speech are brain signals. This study aimed to determine activated brain regions during the imagined speech by spatial and spatiotemporal analysis to assist feature selection in imagined speech recognition. We recorded the normal subjects' brain signals using electroencephalography (EEG) when they imagined saying a word without vocalization. The protocol ran in two different-time sessions to cover the EEG nonstationary. We utilized visual cues in the first sub-session and audio cues in the second one. One word per sub-session had seven trials, each consisting of a rest state and cue presentation, followed by four repetitions of the imagined speech state, each preceded by a preparation state. We calculated power spectral entropy (PSE) values from the acquired dataset and used them as a feature since they captured signals' activity. The normalized PSE above 0.5 (middle value) was considered as high activity. The results showed that repeated-trial protocol induced brain activity during imagined speech. For visual or audio-cued imagined speech, the selected channels were in the central and left frontal regions, with additional right temporal, left parietal, and left occipital regions for visual-cued imagined speech. The highest PSE was found in the central region. Meanwhile, the lowest PSE was shown in the preparation state, confirming that cue presentation was not affecting imagined speech states.