Total: 1
We explore inter-trial, inter-class, and inter-subject variability in covert speech imagination using Electroencephalogram (EEG) signals. Two key functional connectivity metrics (Phase Locking Value and Coherence) revealed unique and shared activation patterns across speech commands, influenced by individual word perception and affective states. We also propose a subject-independent classification model using Hilbert envelope and instantaneous phase features across EEG frequency bands with a Bidirectional Long Short-Term Memory (BiLSTM) architecture, achieving 59.14% classification accuracy across five speech categories. Our state-of-the-art results in EEG-based speech decoding contribute a new understanding of the neural dynamics underlying imagined speech and affective processing.