Generalizable EEG Encoding Models with Naturalistic Audiovisual Stimuli.

Area of Science:

• Neuroscience
• Auditory Processing
• Speech Perception

Background:

• Listeners naturally filter background noise to focus on speech.
• Current speech tracking models often use noise-free or simplified noisy stimuli.
• Generalizability of models from naturalistic environments to controlled stimuli is unclear.

Purpose of the Study:

• To determine if neural responses from naturalistic audiovisual speech stimuli generalize to controlled stimuli.
• To assess the utility of encoding models for naturalistic environments in auditory research.
• To aid data collection from populations sensitive to less engaging stimuli.

Main Methods:

• Recorded noninvasive scalp electroencephalography (EEG) from 17 participants.
• Participants listened to speech in isolation and audiovisual speech with background noise.
• Multivariate temporal receptive field encoding models predicted EEG responses using acoustic, phonological, and visual features.

Main Results:

• Neural responses to naturalistic stimuli were generalizable to controlled datasets.
• Phonological features alone accurately predicted EEG for isolated speech.
• Both phonological and acoustic features improved prediction accuracy for speech in noise.

Conclusions:

• Naturalistic audiovisual stimuli can be used to measure receptive fields comparable to controlled audio-only stimuli.
• Encoding models trained on naturalistic data can reveal auditory tuning generalizable to simpler experimental paradigms.
• Findings support the use of rich, naturalistic stimuli for robust auditory processing research.