EEG Based Imagined Word Recognition Using CNN-Bi-GRU

A notable advancement in brain-computer interface (BCI) for the benefit of the speech-impaired community based on EEG led many insights for imagined speech. This study addresses challenges in translating brain-computer interface (BCI) advancements for speech-impaired individuals using EEG. Existing neural speech decoding methods, despite progress, face classification accuracy issues. The proposed solution combines a deep convolutional neural network (CNN) and a bi-directional gated recurrent unit (Bi-GRU) for improved classification. The systematic approach includes EEG signal preprocessing with bandpass and notch filters, power spectral density analysis, and coherence analysis for cognitive speech task-related EEG topography. The integrated CNN and Bi-GRU enhance memory capacity and classification. Simulation results show the system effectively detects four words with an 88 % recognition rate, surpassing previous performance benchmarks.