Towards Unified Neural Decoding of Perceived, Spoken and Imagined Speech from EEG Signals