A Model-Based Hearing Compensation Method Using a Self-Supervised Framework