Speech causes characteristic neural responses in the brain, both in the cerebral cortex as well as in subcortical areas. Determining relations between speech and the resulting neural activity measured noninvasively through EEG can help to reveal how our brains process the complex information in speech. The recent ICASSP Auditory EEG Signal Processing Grand Challenge therefore set out to explore how segments of EEG data matched and mismatched to a particular speech stimulus can be differentiated best. Our contribution used a combination of slow responses to the speech envelope and faster responses at the fundamental frequency and proceeded to win the challenge.
M. Thornton, D. Mandic, T. Reichenbach,
Decoding envelope and frequency-following EEG responses to continuous speech using deep neural networks,
IEEE O. J. Sign. Proc. 5, 700 (2024) [bioRxiv][pdf]