Fast model-based fitting through active data selection

THE current trend of model-based fitting refers to setting hearing aid tuning parameters (e.g. compression ratios, thresholds and time constants) to values that maximize a utility metric averaged over a representative database. As a rule, a large set of preference data for an individual patient is needed to train his unique utility model. Clearly, this is a situation that is not conducive to an efficient audiology practice. In this paper, we report on a novel approach to accurate model-based fitting that needs few measurements from an individual patient. Our method is based on the observation that, for fitting purposes, we are not interested in an accurate utility model for all possible tuning parameter values. Instead, we are only interested in the values for the tuning parameters that maximize the utility model.