Researchers from the University of Pittsburgh and Northeastern University have developed an approach to improve auditory processing in hearing-aid technology to provide more realistic sounds to users, improving communication and learning ability.
Description
In noisy environments, for example, at a party a listener can usually focus on one voice of interest, selectively hearing what is being said and reducing background noise. This auditory attention detection (AAD) uses complex neural networks, the auditory attention network. In those with hearing impairments or hearing loss, this is a significant challenge impacting on communication and learning abilities. This novel system aims to mimic the cortical processing of the auditory attention network using state-of-the-art AAD technology.
Applications
• Smart hearing aids
• Brain-embedded device interfaces (BECI)
Advantages
Approximately 35 million Americans suffer from hearing loss. While hearing aids can help, they are less helpful in high noise situations and fail to hone in on one voice and dampen the sounds of irrelevant background noise. These limitations have implications on the user’s ability to communicate or learn and work in crowded environments. Current approaches to AAD make use of statistical or artificial neural network (ANN) architecture. However, these statistical models are not practical for use in BECI as they require longer decision windows and use many EEG channels. ANN also uses high amounts of power.
In non-hearing-impaired people, the hybrid convolutional neural network-spiking neural network (CNN-SNN) architecture utilizes electroencephalogram (EEG) signals to detect the auditory attention of a listener. This device utilizes CNN-SNN architecture to overcome the existing challenges in AAD technology, allowing for the development of practical BECI applications with real-time auditory attention tracking, requiring fewer EEG channels, lower power consumption and with a lower memory requirement.
Invention Readiness
A device has been developed using CNN-SNN architecture, inspired by the auditory cortex using only 8 EEG electrodes, strategically placed close to the auditory cortex. It has been shown to successfully decode auditory attention with a latency of 1 second with an accuracy of over 91%, significantly higher than existing technology.
IP Status
https://patents.google.com/patent/WO2024107619A1
