Bayesian simulation of hearing thresholds at conventional and extended high frequencies in young adults

The goal of this study was to characterize the population-level distributions of hearing thresholds at conventional and extended high frequencies (EHFs). To achieve this goal, distributions of typical hearing thresholds expressed in dB SPL were measured in an otologically normal young-adult sample. Thresholds were obtained using an adaptive procedure with the Sennheiser HDA-200 earphone. Thresholds were measured at 16 frequencies, ranging from 250 Hz to 16 kHz, in a non-random sample of 126 young adults (18-36 years old) with typical hearing (≤ 20 dB HL at conventional frequencies). Using the posterior of a Bayesian multilevel distributional regression model that was fit to the data, we simulated hearing thresholds at conventional and extended high frequencies for 18,000 new participants. The simulated thresholds were used to calculate probabilistic estimates of the reference equivalent threshold sound pressure levels (RETSPLs) and other percentiles of the distributions of hearing thresholds at each frequency. We used these percentile estimates to define the range of typical hearing sensitivity in dB SPL and found that this range varies across frequency, unlike the current fixed 20-25 dB HL cutoffs that are used to classify "normal" hearing. Specifically, at conventional frequencies, the fixed 20-25 dB HL cutoffs overestimated the range of typical thresholds, whereas at extended high frequencies, the fixed cutoffs overestimated this range. The proposed probabilistic approach provides a framework for characterizing the information from pure tone thresholds in terms of the range of hearing sensitivity, rather than categorizing hearing ability on the basis of a frequency-invariant criterion.