Temporal variation in the acoustic dynamic range is a confounding factor in EEG-based tracking of absolute auditory attention to speech

Abstract

Many studies have demonstrated that auditory attention to natural speech can be decoded from EEG data. However, most studies focus on selective auditory attention decoding (sAAD) with competing speakers, while the dynamics of absolute auditory attention decoding (aAAD) to a single target remains underexplored. The goal of aAAD is to measure the degree of attention to a single speaker, has applications for objective measurements of attention in psychological and educational contexts. To investigate this aAAD paradigm, we designed an experiment where subjects listened to a video lecture under varying attentive conditions. We trained neural decoders to reconstruct the speech envelope from EEG in the baseline attentive condition and use the correlation coefficient between the decoded and real speech envelope as a metric for attention to the speech. Our analysis shows that the envelope standard deviation (SD) of the speech envelope in the 1-4 Hz band strongly correlates with this metric across different segments of the speech stimulus. However, this correlation weakens in the 0.1-4 Hz band, where the degree of separation between the attentive and inattentive state becomes more pronounced. This highlights the unique contribution of the 0.1-1 Hz range, which enhances the distinction of attentional states and remains less affected by confounding factors such as the time-varying dynamic range of the speech envelope.

Simon Geirnaert

Postdoctoral researcher

My research interests include signal processing algorithm design for multi-channel biomedical sensor arrays (e.g., electroencephalography) with applications in attention decoding for brain-computer interfaces.