Decoding selective auditory attention from continuous speech: A TRF-based approach to dichotic listening paradigms

Selective auditory attention allows us to focus on a single conversation in a crowded room. Traditional dichotic listening paradigms for brain activity studies rely on probe tones embedded in audio to elicit event-related potentials (ERPs). In this setup, participants attend to speech in one ear while ignoring the other, using the ERP difference across ears as an index of selective attention. While effective, this approach requires artificial stimulus manipulation, constraining experimental design and ecological validity. Here we compare ERP and Temporal Response Function (TRF) analyses. We show that TRFs fitted to the continuous speech signal recover the better attentional modulation captured by probe-tone ERPs, without relying on artificial stimuli. TRFs make it possible to characterize, with fine temporal resolution and channel-level specificity, how the auditory system differentially encodes attended versus unattended speech, all while preserving naturalistic listening conditions. Moreover, it allows us to discriminate between speech attributes that could differentially separate those conditions. We discuss the advantages of this approach for studying selective attention in more ecologically valid settings. We present a new Python library (TRFlearn) that covers the whole M/EEG analysis pipeline in a wide range of naturalistic settings. Our results suggest that TRFs offer a powerful, flexible alternative to ERP-based methods in dichotic listening research.