Silent Speech Recognition based on sEMG and EEG Signals

In recent years, silent speech recognition attracts attention with particular interests either using surface electromyogram (sEMG) or electroencephalogram (EEG), but both have limitations. This paper proposes to combine sEMG and EEG signals for recognizing silent speech. Data are first preprocessed to get clean signals followed by extracting sEMG and EEG respectively. Then time domain features and entropy features are separately extracted from the dataset, which are classified by a typical random forest classifier. Feature fusion and decision fusion are also researched to explore their effects on recognition. Our experimental results show that entropy features play a certain role in recognition, and the second level decision fusion obtains a better result with a correct rate of 86.53%. It may remind more researchers of combining EEG and sEMG in this field.