This short paper summarizes ongoing work on a probabilistic modeling approach to the design of personalized hearing aid algorithms. In this framework, we first specify a probabilistic generative model that includes an explicit description of the hearing loss problem. Given the model, hearing aid signal processing relates to on-line Bayesian state estimation (similar to Kalman filtering). Estimation of the tuning parameters (known as the ‘fitting’ task in hearing aid parlance) corresponds to Bayesian parameter estimation. The innovative aspect of the framework is that both the signal processing and fitting tasks can be automatically inferred from the probabilistic model in conjunction with patient appraisals (the data).