THE NEURO-COMPENSATOR: A HEARING COMPENSATION ALGORITHM BASED ON COCHLEAR MODELING AND MACHINE LEARNING
 
Becker S (1), Pango P (2), De Laat JAPM (3).
 
(1) McMaster University of Toronto, Canada; (2) VitaSound Audio, Ontario, Canada; (3) Leiden University Medical Center, Netherlands.

In this presentation, we will describe the design and testing results of a novel hearing aid based on the Neuro-Compensator algorithm. The Neuro-Compensator employs neurocomputational models of the pathophysiology of hearing loss to optimize the gain calculation in a hearing aid so as to generate a more normal pattern of neuronal firing in the auditory nerve of the hearing impaired listener.

The Neuro-Compensator employs sophisticated computer models of the cochlea to determine the optimal gain calculation for a given hearing impaired listener. The auditory models are used to simulate how the auditory nerve of a hearing impaired person would respond to a given level of amplification across all frequency channels. The resultant auditory neuronal firing pattern for the impaired ear model is compared to that of a healthy ear model. Machine learning methods are used to iteratively adjust the gain calculation performed by the Neuro-Compensator, until a normal pattern of neuronal activity is restored. Once this training procedure is completed, the Neuro-Compensator no longer requires the complex auditory model, and the final gain calculation can then be incorporated into a conventional hearing aid microprocessor. In the resulting hearing aid, rather than calculating the gain separately within each frequency channel, the gain of each channel is dynamically calculated as a function of the entire spectral content of the signal. The Neuro-Compensator thereby has the ability to restore the non-linearities and cross-channel modulatory functions normally achieved by the outer hair cells. Computer simulations indicate that compared to the widely used WDRC gain calculation, the Neuro-Compensator is better able to restore intelligibility of higher frequency components of speech signals. Preliminary subjective reports from hearing-impaired individuals indicate that compared to conventional hearing aids the Neuro-Compensator-based hearing aid restores a much more natural sound for both speech and music, while sound perception in noise and sound localization are greatly enhanced.

E-mail: japmdelaat@lumc.nl