From business need to model monitoring: Optimizing manufacturing of custom hearing aids with machine learning

Authors: Anthea Bott1

1GN Hearing

Background: Hearing aids must pass a series of electroacoustic quality tests – known as a digital signal analysis (DSA) – before they can be sold. In-the-ear style hearing aids (custom hearing aids) are handmade and are therefore slightly more prone to DSA test failures as compared to behind-the-ear style hearing aids which utilize automation. Reasons for DSA failures include but are not limited to loud environmental noise when testing, poor position of the hearing aid in the test chamber, component failure, and feedback. Depending on the reason for the failure, the hearing aid should be retested or sent to be reworked; however, manufacturing staff are not provided with DSA failure reasons. Our internal investigation found that feedback was the most common cause of DSA failures in custom hearing aids and that these devices must be reworked before they pass DSA testing. The aim of this project was to alert manufacturing staff to DSA failures on custom hearing aids due to feedback so the hearing aid could immediately be sent to be re-worked.

Method: A five-step machine learning life-cycle approach was utilized. Steps included defining the business case, data exploration, model development, model deployment, and model monitoring.

Results: This presentation will discuss the five stages of the machine learning life-cycle approach utilized in this project, with a particular focus on cloud-based model deployment and monitoring. We will describe a number of challenges and learnings and considerations for future work.