EEG-Based Identification of Native and Non-Native Language Perception Using Deep Learning

This work classifies the identification of native and non-native language speakers using the electroencephalogram (EEG) signals and Deep Learning techniques in native and non-native musical songs. The EEG based dataset is recorded from 20 different subjects in experiments of native and non-native language stimuli, using two different modalities (a) in-ear headphones, and (b) bone-conducting headphones. For the Classification of native and non-native speakers on EEG responses, deep learning models including Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), CNN-LSTM, Bidirectional LSTM, and Dense Neural Network (DNN) are implemented. Experimental results demonstrate that the Bidirectional-LSTM model achieved the highest accuracy (98%). This high accuracy validates the usefulness of the proposed system in real world applications, such as by intelligence agencies, to accurately identify understandable communication means of suspects under investigation.