The Learning Prescription, A Neural Network Hearing Aid Core

Cite this article

@article{Flax-TLP-2023,
  author  = "Matt R. Flax",
  title   = "The Learning prescription, A Neural Network Hearing Aid Core",
  journal = "Flatmax Pty Ltd",
  year    = 2023,
}

Abstract

The definition of a hearing aid core which is based on a prescription neural network (such as NAL-NL2) is defined here. This hearing aid core replaces a traditional compressor hearing aid core which mimics the said hearing aid prescription. Whilst the replacement of the compressors for a neural network may seem simple, the implications are vast in terms of the “learning prescription” where the topology of the neural network may be increased to make available more free parameters and allow great personalisation of the hearing aid prescription.

History

Since around 2011 I have been trying to walk this concept of replacing the core of the hearing aid with a neural network which would allow it to continuously learn away from a static prescription according to the user's preference. While the original NAL-NL2 presciption algorithm used a neural network to prescribe gain matrices, it wasn't suitable for generating personal neural networks. Since 2012 I made that possible.

I have walked this concept privately through the top level of institutions like NAL (with two different heads of research, Starkey (for their variety of prescriptions) and others. I have also tried to interest entrepreneurs and others. Back twelve years ago (it is currently 2023), people didn’t really understand the vision of rebuilding hearing aids to be neural processors. Nowadays with the advent of chatGPT and other “AI” systems, people are starting to understand the power of neural networks and machine learning in general. The concept of layering and redesigning to have desired effects - beyond the learning prescription.

I have had the working NAL-NL2 prescription variant which generates hearing aid prescriptions to run on the type of system specified by this document since 2012.

Why is this article and code on github ?

This content is now date stamped and in the public domain. A preprint server could be used to date stamp the article, however preprint servers can't manage software code and have no mechanism for integrating feedback from the community. By using github (or a differnet code management system like gitlab), people who want to review this document and code can do so by opening issues (top left button "Issues" on the github webpage). People who want to help by bug fixing or even in some cases contrubuting can also do so by forking this repository and submitting pull requests (PRs).