Towards hearing-aid personalization: preference elicitation from audiological data

Given a sound library, a sound sample and two parameter settings are selected to generate two hearing-aid output signals. The patient gives a preference response, whether he prefers the first or the second signal. Patient preferences are modeled by a utility function. Our uncertainty about the “true” utility model of a patient is represented by a probability distribution which is updated using Bayes’ rule after each listening experiment. Using the new probability distribution, this process is repeated.