Introduction: Pupil dilation can serve as a measure of auditory attention and as an additional measure of hearing threshold. Studies in infants and adults show a difference in responses to speech and other sounds. It is unknown, however, how much exposure is needed to reliably observe this difference at a comfortable levels of intensity, how reliable is the measure of pupil diameter response (PDR) in individuals at various intensity levels, and whether we can observe systematic differences in the response to the specific type of deviant sounds.

Methods: We observed the PDR to tones and speech (ling-6-sounds) stimuli during passive listening at different intensities in two groups of young adults (N = 24, ME = 29 years, DS = 3.9, 11 females). An oddball paradigm with 20% of deviant sounds was used in both experiments. The time windows, in which the presence of a deviant sound elicited PDR compared to the standard sound across different intensity levels, were estimated by computing the cluster-based statistic using the permuted likelihood ratio tests. The averaged values of these time windows were used to model the group responses and predict individual performance.

Results: In both groups, the augmented PDR was associated with deviant sound stimuli. At the highest tested intensity level (70 dB, reported as comfortable by all participants), the analysis of 10 deviant and 10 standard trials (but not smaller amount of data) yielded reliable model predictions (tones: sensitivity = 0.83; sensitivity = 0.75, positive-predictive-value (PPV) = 0.77; speech: sensitivity = 0.83; sensitivity = 0.5, PPV = 0.63). Averaged raw data per participant yielded even higher PPVs (0.92 and 0.83). Further analysis revealed that in the tone experiment, only high frequency deviant tones (2 & 4 kHz) elicited significant change in PDR, whereas in the speech experiment, consonants (/ss/ and /sh/) but not vowels (/i/, /u/) elicited significant change in PDR.

Discussion: In this study, the minimal amount of exposure to tone and speech stimuli at the comfortable hearing level needed to fit a regression model and to reliably predict the performance in individual participants was measured. This represents the necessary step in creating the PDR-based adaptive procedure with which auditory attention can be measured. We also show that the PDR does not only depend on the type of the sound (speech, noise, tones) but also on the internal categories (e.g. vowels vs voiceless consonants).