Authors: Ines Ben Aïssa¹, Rim Amara², Mériem Jaïdane³
¹L3S-ENIT, National Engineering School of Tunis, Tunisia; ²INSAT & L3S-ENIT, Tunis, Tunisia; ³dB.Sense & L3S-ENIT, Tunis, Tunisia
Background: Conventional audiometric testing, such as the Hughson-Westlake (HW) method, often feels long and repetitive to patients. This study investigates the difference between perceived and objective test duration by comparing HW with a Bayesian Active Learning (BAL) approach. Musicians were chosen as a target population due to their high pitch discrimination abilities, which minimize inter-individual variability and highlight fundamental perceptual mechanisms.
Method: Six young musicians underwent both BAL and HW assessments, separated by a music listening session. BAL utilized Gaussian Process inference to dynamically adapt stimuli across a continuous frequency range (125-8000 Hz), while HW followed the standard “down 10 dB / up 5 dB” rule at nine fixed frequencies. Participants reported on fatigue and perceived duration via post-session questionnaires. The predictability of the stimulus sequences was formalized using the maximum Lyapunov exponent as a metric of the “predictability horizon.”
Results: Median test duration was significantly shorter for BAL (5:38 min) compared to HW (15:52 min). While BAL participants accurately estimated the test duration (median ratio 1.1), HW participants substantially underestimated it (ratio 2.3). This underestimation in the HW group is consistent with predictability-driven temporal distortion; the structured, repetitive nature of the HW protocol engages anticipation mechanisms that compress the perception of time. In contrast, BAL provides a more randomized, and thus accurately perceived, stimulus structure.
Conclusion: BAL audiometry not only increases patient throughput by reconstructing a continuous audiogram in less time but also provides a more accurate subjective experience of duration. The use of dynamical systems theory to quantify stimulus predictability offers a new framework for optimizing patient engagement and reducing the cognitive biases associated with conventional audiometric procedures.
