Tracing and decoding of covert phonemes using single channel Electroencephalogram with Machine Learning Techniques

Abstract

A Brain-computer interface BCI is a technology that interfaces the brain and computer for communication without the person expressing it. Amongst concepts of reading thoughts of the brain, decoding covert speech is a popular application in BCI which can be able to translate the imagined voice inside a person. In this study, Electroencephalogram (EEGs) has been used to interpret the covert speech of a person. On the other hand, reading the brain with EEG is a complicated task to use in daily life applications as it needs multichannel spatial information to be extracted by connecting leads all over the scalp. In the direction of overcoming this complexity, this study uses only single-channel EEG Fpz, which is much easier to access than channels. In this study, Multilayer Perceptron (MLP), K-nearest neighbour Classifier (KNN), Support Vector Classifier (SVC), and Random Forest (RF) models are proposed to classify a single channel Fpz of EEG by extracting spectral information in form of wavelet decomposition coefficients and an energy level over Alpha, Beta, Gamma, Delta and Theta bands to show the evidence that covert speech can be derived through single channel EEG with basics classifiers.