OPTIMIZATION OF PRE-PROCESSING ROUTINES IN SPEECH IMAGERY-BASED EEG SIGNALS

Abstract

Speech imagery is one type of mental imagery specific to processing verbal sequences and plays a vital role in human thought processes. Speech imagery has become an interesting paradigm for researchers as speech imagery has a high similarity to real voice communication. Electroencephalography (EEG) is a noninvasive electrophysiological technique that measures the mental state of the brain directly from the scalp. The nature of the acquired EEG signals is nonlinear and nonstationary. As EEG signals have a low signal-to-noise ratio (SNR), artifacts occur during acquisition. Hence, an efficient framework of pre-processing is required to obtain artifact-free EEG for further applications. Selection of the optimal pre-processing techniques for EEG still remains a challenging task. This work focuses on employing and comparing the different pre-processing techniques and lists out the optimal solutions for pre-processing Speech imagery-based EEG signals. The techniques are compared based on the Mean Square Error and Peak Signal-to-Noise Ratio values.