Hearing aids are the principal tool today for ameliorating age-related hearing loss and its significant social, cognitive and functional costs to patients and society at large. However, many individuals who are prescribed hearing aids do not use them at all, or use them only occasionally. Most reasons behind the ?hearing aid in the drawer? phenomenon relate to the characteristics of the sound produced, and could, in theory, be addressed with the correct signal processing strategy. The problem persists despite the increased complexity and power of new devices, for three reasons: (a) The hearing aid parameters, as set in the clinic, introduce distortion or render audible many sounds that the hearing impaired user had become accustomed to not hearing. The novelty is often so uncomfortable for the user as to discard the device. (b) The optimum parameters vary depending on the listening task and environment. Under some conditions, a device with parameters designed for a different condition will perform worse than no device at all. (c) The clinical fitting is derived from a non-ideal way to assess auditory function (the pure- tone audiogram). The optimum parameters for the actual impairment may be different from those of the prescribed fitting. Although it is true that the physiological mechanisms make it impossible to process sound so as to completely reverse the effect of sensorineural hearing loss, a device that delivers some benefit at all times is likely to be used all the time. The goal is to develop a hearing aid that can adaptively change its parameters to address the problems above, and will be accomplished with a novel fitting approach that rapidly presents a number of parameter settings to the user and lets the user guide the system toward the optimal settings for each listening situation. This requires the development of machine-learning algorithms to effectively search the parameter space and user interface devices and instructions that are easy for the patient to use. The focus of this Phase I proposal is the development of the algorithms and the adaptive user-driven fitting program, and to compare the proposed fitting with the traditional audiogram-based fitting across measures of functional hearing (ability to recognize speech in noise) and subjective preference.
A hearing aid user is often dissatisfied with the sound quality of their device, despite its sophistication and adjustment by a trained audiologist. The problem can be mitigated by letting the user fine-tune the device for maximum comfort in everyday use. We will apply modern machine learning methods to develop a program for efficient user-driven fitting of hearing assistive devices.
