Neural correlates of environmental noise soundscapes : an EEG study

Environmental noise has documented effects on productivity in the workplace, and suggested impacts on health and wellbeing. However, there remains a gap in knowledge in determining whether there are neural markers for these effects that might be used in design, planning, and stakeholder engagement. Neuro-physiological measurement has become practical in laboratory listening tests, due to advances in in dry electrode technology, fast analogue-to-digital conversion, and cross-platform synchronisation, allowing for simultaneous ambisonic playback and collection of listener response data in multimodal contexts. The datasets created by such measurement are large and typically impractical to analyse over significant numbers of trials without modelling. In this work we present results from a pilot study (number of participants N=37), in which listeners were exposed to a randomised playback of first-order ambisonic recordings of typical urban environmental soundscapes (aircraft, trains, road traffic, and construction noise). Electroencephalograph (EEG) measurements were captured synchronously across a 10/20 scalp position. Data for each subject was normalised and smoothed before being filtered into alpha and beta frequency bands using PSD calculations, before being further filtered to remove artefacts including high frequency interference and event-related potential activity such as blinking and similar head movement. Self-reported data on perceived annoyance was also captured using the ISO 15666 scale from each participant in response to the stimulus set. We subsequently extract three acoustic components across the stimulus set using signal processing analysis techniques; loudness, sharpness (as a factor of spectral centroid), and mel-frequency cepstral coefficients (MFCC), and map these against neural activity indicated by correlates in the EEG recordings. We also compare EEG recordings with self-reported levels of annoyance. We plan further work to train a regression model with weighted vectors for EEG activity, acoustic features, and self-reported annoyance.