Self-fitting based on speech MCL measurements and Bayesian optimization methods

In clinical settings, hearing-aid gain is typically prescribed based on pure-tone audiograms. Outside the clinic, audiometric testing is complicated by problems with background noise and calibration. A promising candidate for out-of-clinic applications is the most comfortable level (MCL) for prolonged speech listening. However, existing methods for conducting MCL measurements are unsuitable for determining frequency-specific amplification. This study investigated a new method for measuring speech MCL based on pairwise comparisons and Gaussian process models. The method was then used to estimate individually preferred, frequency-dependent gain (IPG). Fifteen normal-hearing and 13 hearing-impaired participants were tested. In Part 1, the new MCL method was compared to a standard bracketing method in terms of precision and time efficiency. In Part 2, IPG settings were determined using candidate gain settings calculated based on the NAL-NL2 fitting rule and a set of standard audiograms. The participants made pairwise comparisons of the candidate gain settings and indicated their preference in terms of listening comfort and speech clarity. Individual NAL-NL2-based reference settings were also calculated. The bracketing and new method were similar in terms of precision and time efficiency. Further, the new method returned IPG settings that were sensitive to differences in listening comfort and speech clarity. For some, but not all, participants, the IPG settings aligned with the NAL-NL2 reference settings.