Effects of personalizing hearing-aid parameter settings using a real-time machine-learning approach

In most hearing-aid fittings, amplification is prescribed by a fitting rationale that uses the audiogram as the main input. This approach may fail in situations where the user’s listening intention deviates from that assumed by the rationale. This shortcoming motivated a new commercially available method to self-adjust hearing-aid parameters while in a specific situation. The method is based on machine-learning algorithms that estimate the setting that optimizes user satisfaction based on user preferences in paired comparisons of parameter settings. We present results from a lab study where 20 participants with hearing loss used the method to adjust hearing-aid gain in 12 different sound scenarios with respect to three different sound attributes, and subsequently, in a double-blind assessment, compared the adjusted settings with the prescribed settings. The results showed a benefit of the method on basic audio quality. A large spread in the gain adjustments was observed, suggesting the need for more personalized settings of hearing aids. We also present anonymous user data gathered during real-life use of the method, which indicate when and why the method is used. We compare these data to general investigations of listeners’ auditory reality and suggest clinical and rehabilitative implications of using the method.