Research Overview
The Australian Future Hearing Initiative (AFHI) is a three-year project, announced in March 2023, that brings together Google and world-leading organizations in hearing accessibility, including Cochlear and other members of the Australian Hearing Hub. The initiative addresses a key challenge: while current hearing aids work well in quiet places, they often fail to provide adequate support in complex, noisy social environments. With 1.5 billion people affected by hearing loss globally—a condition costing an estimated $1 trillion (US) annually—the AFHI focuses on applying new AI and machine learning (ML) technologies to develop listening and communication devices that offer highly customized hearing healthcare.1
The first workstream, “Hyper-Personalisation,” aims to develop hearing solutions that compensate for the specific causes of a person’s hearing loss, rather than just managing the symptom of reduced sensitivity. The team is using an innovative approach involving a sophisticated computational model of the inner ear called CARFAC to simulate both normal hearing and various types of hearing impairment. By analyzing the differences between these two model outputs, they are training ML models for fitting traditional hearing aids, training an innovative ML hearing aid and exploring new coding strategies for cochlear implants. In addition, new hearing tests have been developed for on-line delivery that examine traditional measures of hearing function (audiometric sensitivity) as well as new tests of hearing function at sound level relevant for normal conversational interaction (suprathreshold tests). Clinical testing of the new hearing performance tests and the ML based fitting of standard hearing aids is progressing and testing of the ML hearing aid should begin in coming months.
The second workstream, “PRISM (PRedicting Interaction difficulty using Sensors and Micro-EMA)”, seeks to bridge the gap between objective lab tests and the actual user experience in real-life settings. PRISM is developing a portable platform that uses a combination of smart watch-based physiological sensors (like those tracking heart rate and stress) and machine learning-based acoustic analysis to understand and ultimately predict a listener’s difficulty and listening effort in real time. In the first instance, this enables a measure of the effectiveness of different hearing aid algorithms and technology under realistic conditions of use. Prediction of listening effort can also then be used as a control signal to personalize and enhance the hearing device’s function. The smart phone app for data collection including integration with the smart watch, the web portal for setting up and managing experiments in the wild and the cloud storage back end have all been completed and are undergoing testing before open sourced release on GitHub.
The third workstream, “AFHI Impact” addresses two major issues: (i) hearing testing requires a specialist audiologist and is not easily accessible to many people and (ii) over 430 million people worldwide—80% in low-income countries—lack access to affordable solutions for disabling hearing loss. To date, AFHI has made significant technical progress, including developing a prototype machine-learning hearing aid and creating new online tests for hearing performance. The “AFHI Impact” workstream focuses on three areas to solve these issues: promoting the open-sourced tools and outcomes to scientific, media, and government communities; raising public awareness, providing education on hearing loss prevention and promoting the on-line hearing test; and developing new models for global access, specifically by investigating ways to facilitate the production of very low-cost, high-quality hearing aids for people in low-resource countries.