Comparative Analysis of EEG-Based Sound Location Decoding between Real and Virtual Listening

Authors: Nils Marggraf-Turley1

1Imperial College London

Background: Neural responses to real and virtual acoustic listening have been extensively studied, elucidating cortical representation of spatial auditory perception. Yet, differences in neural responses between spatialisation methods remain underexplored. Head-Related Transfer Functions (HRTFs) are essential for creating virtual auditory environments but diminish spatial perception when non-individual. Previous research demonstrated the ability to decode sound source locations from Event-Related Potentials (ERP) from free-field listening. We asked whether decoding is feasible from non-individual HRTFs (virtual) listening and investigated differences in decoding accuracy (DA) to free-field (real) listening.

Method: Participants were exposed to sounds from four azimuthal locations, symmetric around interaural and median planes, via loudspeakers or earphones. ERPs were recorded, and a classifier was trained to discern between locations from the entire response or individual time points. Behavioural localisations were also recorded.

Results: Significant DA was achieved for all subjects for median and interaural-plane-symmetric decoding from real listening. In contrast, significant decoding for all subjects occurred only for median-plane-symmetric locations from virtual listening. Significant differences in DA for interaural-plane-symmetric positions were found between real and virtual conditions, and a qualitative difference in peak DA latency was identified. Behaviourally, most front-back confusions occurred during virtual listening, with weak correlation between localisation accuracy and DA.

Conclusion: Lower DA for interaural-symmetric virtual locations suggests weaker neural representations of spectral cues under non-individual HRTF listening. Behavioural trends suggest that these DA differences may reflect localisation accuracy. DA latency suggests cortical processing delays between real and virtual contexts supporting analogous research findings.