Impact of Fragmentation on Temporal Event Localization for Speech EEG

This study investigates the effect of temporal event localization of speech-imagery in electroencephalogram (EEG) signals. Temporal event localization is required for speech-imagery EEG signals as they are often recorded over long segments with the actual event being a fraction of the recorded duration. It is achieved by fragmenting the EEG epochs into smaller components and identifying those fragments from different signals that are common to the same class. After extracting these fragments and merging them into a new set of signals, they are validated using a set of classifiers. The results show a significant improvement in the classifier performance when compared to using signals without fragmentation. Extracting these fragments would also allow for extracting features of syllables (and eventually, phonemes), as the same algorithm used to fragment the signals may be used to extract features of those word components.