| Popis: |
One of the main complaints of hearing aid users is the difficulty of and need for frequent adjustments they have to make to the device. Typically, volume control and other settings prescribed during the hearing aid fitting process still need to be fine-tuned in real life to meet the specific varying needs of the user in different environments. Recently, learning algorithms have been introduced to minimize these problems. By taking a weighted average of the past adjustments, the device is able to adapt to the preferred setting for each environment presented. Typically, fixed learning algorithms are used for all users and environments. In reality, users have different behaviors and the optimum learning time is expected to vary over users and environments. In this thesis, we study the potential for user-specific adaptive learning of the user settings. Profiles of various types of user behaviors were generated and the optimum time constant for learning user preferences was determined in each case using fixed exponential smoothing. We show that the performance of the algorithm is clearly dependent on the user profile and no single fixed time constant is optimum for all users or situations. Three adaptive exponential smoothing methods were then evaluated. Four performance measures are proposed to evaluate these methods based on the users' preferred setting when they re-enter the same environment. For some user profiles, adaptive methods were found to perform as well as the optimum time constant when learning user preferences for volume control without having prior knowledge of the users' behavior. In particular it is shown that the method by W. M. Chow ( Journal of Industrial Engineering, 16, pp. 314-317, 1965) can be tuned to perform as well as the optimum time constant for a given user. |
