AUDIOCAP
Main issues and objectives
Context
The main complaint of adults with hearing loss is difficulty understanding speech in noise. Restoring intelligibility in noise for the elderly is now a public health issue, due to the increase in social isolation, depression and dementia among patients without hearing aids. Hearing aids currently play a beneficial role in preventing cognitive over-decline, but one of the obstacles to their fitting is the lack of efficiency in noisy environments.
From a medical point of view, patients are fitted by an audioprosthetist, whose role is also to adjust the prosthesis. This adjustment process is very long and requires several patient visits, generally several weeks apart.
From a technical standpoint, the hearing aids currently on the market cannot integrate high-performance processing in noisy environments – such as denoising algorithms based on the use of artificial intelligence (e.g., using deep neural networks) because of the limited resources, in terms of energy and computing power, of the current devices.
The project is structured around 4 batches, and IRIT is involved in the first 2:
- Prediction of intelligibility
- Improvement of intelligibility
- Comportemental tests
- User integration
Prediction of intelligibility
The aim of this batch is to create models for automatic speech recognition in noise, in order to simulate the recognition rate of hearing impaired patients. These models will help speed up the process of hearing aid fitting by the hearing care professional.
Improvement of intelligibility
The aim of this batch is to develop denoising algorithms based on signal processing methods in order to improve intelligibility.
Partners
- ARCHEAN Technologies
- DSI
- UPS – IRIT – SAMoVA
- CHU Toulouse
- Ecole d’audioprothésiste de Cahors
People involved in the SAMoVA team
- Julien Pinquier
- Verdiana De Fino
Funding
- Région Occitanie dans le cadre de l’appel READYNOV
Schedule
- Start time: 1st January 2019
- End time: 31st December 2022