Fraunhofer Institute for Digital Media Technology IDMT
More than half of adults over the age of 60 suffer from hearing loss (World Report on Hearing, World Health Organization, 2021). Those affected often do not seek hearing assistance, which increases the risk of further damage, such as accidents at work, depression, or dementia.
The aim of the AIHEARS project is to research, develop, and evaluate an AI-based hearing system that takes individual hearing needs and preferences into account and automatically applies them in everyday listening situations. The consortium wants to help make hearing care easily accessible and facilitate independent adjustment of hearing solutions to individual needs. Within a hardware-independent application, common medical hearing aids in the form of hearing devices as well as non-medical devices, such as headphones or earbuds, should be taken into account. The aim is to promote the lowest possible threshold for use and early adoption of hearing amplification.
The AI-based control system in AIHEARS is designed to be trained by users themselves in their individual daily listening environments. Everyday listening environments are classified and sound adjustments are assigned. The trained algorithm will then be able to predict individual sound adjustment preferences in real time and apply them in automatically recognized listening environments.
Fraunhofer IDMT's contribution to the project
Fraunhofer IDMT-HSA is responsible for researching customizable audio signal processing in the project. This includes, in particular, the development and evaluation of algorithms for customizable, selective hearing through intelligent source separation. Among other things, these algorithms are designed to ensure that certain relevant sounds are extracted from a mix of different speakers and background noise. Corresponding personalization concepts are intended to ensure that the individual sound preferences of users are automatically taken into account in everyday life. To achieve this, the developments will be evaluated several times during the course of the project and optimized for real-time operation using models and in listening tests.