Personalization of Hearing Aids through Bayesian Preference Elicitation

Modern hearing aids have evolved into complex digital signal processing (DSP)
devices. The tuning of DSP parameters so as to maximize user satisfaction, i.e.
fitting of a hearing aid, has become similarly complex. Several factors
contribute to the complexity of this optimization task, including:
* the dimensionality of the parameter space is large;
* the determinants of hearing-impaired user satisfaction are unknown;
* the evaluation of this satisfaction through listening tests is costly (in
terms of patient burden and clinical time investment) and unreliable (due to
inconsistent responses).
As a result, the vast majority of the hundreds of parameters of a hearing aid
are fixed by the manufacturer at an average value. Fewer than ten of the
parameters are personalized by the hearing-aid dispenser based on the severity
of the hearing loss. Thus, fitting of hearing aids relies more on empirical
procedures than on mathematical optimization methods.

The goal of the HearClip project is to improve the hearing-aid fitting
procedure. The development is tweaked by listening experiments.

This would be easy, if we knew what determines user satisfaction. Since we do
not know this, within HearClip we model the uncertainty about user satisfaction
through probability distributions. The framework is Bayesian incremental
utility elicitation. To apply this framework to hearing-aid fitting, it is
planned to develop algorithms that are computable in a reasonable amount of
time. This involves research into numerical issues of approximate inference in
probabilistic models and efficient optimization in large domains. Another part
concerns the development of probabilistic models for user satisfaction in the
hearing domain and the construction of sound libraries.

The current experiment involves the evaluation of (parts of) the newly
developed framework, as well the clinical evaluation of the developed
procedure. Each subject is his/her own control, by comparing the results of
different hearing-aid settings to other setting for the same subject. The
evaluation of the entire fitting procedure will be done by comparing the
results to a conventional fitting method, if available.

For each subject, testing time will not exceed 3 hours per session. All
experiments will be limited to at most two sessions. The experiment has,
besides the time spent, no adverse consequences. There are no risks associated
with participation.