The “cocktail party effect” is the term researchers use to describe the brain’s ability to pick out one person’s voice from a multitude of voices and sounds. However, this ability to hear selectively declines significantly with age, says Li. “Aging people lose the natural ability for selective auditory attention, and that is not just because their ears are failing, but because their brains are changing too.” When an older person takes a hearing test, they often receive expensive hearing aids afterwards. “However, there is a massive point of frustration with the hearing aids. Seniors still struggle to understand speech in a noisy room because traditional hearing aids amplify everything – background chatter is now just as loud as the foreground voice.” Thus, while today’s hearing aids are able to compensate for the declining hearing ability of the ears, they can do nothing against the loss of selective auditory attention in the brain, according to Li.

Electrodes as a Connection Between Brain and AI

It is possible to decode who a person is currently trying to listen to – even in older adults – using brain wave data. These brain waves are measured using electroencephalography, or EEG for short. But how can the corresponding information be transmitted to the hearing aid and decoded and processed there using AI so that the currently relevant voice is selectively extracted and amplified? Haizhou Li has intensively researched this core question of human-machine interaction as a U Bremen Excellence Chair holder over the past seven years. Excellence Chairs are awarded to renowned researchers who act as bridges between the University of Bremen and the world’s leading universities. Li is a leading computer scientist in the field of speech processing and AI from the National University of Singapore.

Center Stage of Scientific Discovery

When he was appointed U Bremen Excellence Chair holder in 2019, he had just begun research on neuro-steered auditory attention. “The U Bremen Excellence Chairs program was not only a stepping stone for my new exploration but also the center stage of scientific discovery for the past seven years,” he says. “Through the program, the collaboration between Singapore and Bremen has flourished to become a reputable center of excellence in the field.”

The most challenging part of this research was “to bridge the gap between two completely different functional systems – a biological brain and an artificial intelligence system,” he says. This connection is to be formed in the future by small EEG electrodes on the hearing aid. AI-based technology in the device can use the EEG data obtained to recognize which voice the listener’s attention is focused on and selectively amplify the relevant voice in real time, according to Li.

This world-first neuro-steered hearing system could revolutionize the development of future hearing aids in the coming years. After all, it would compensate for the lost selection ability of the brain in dealing with voices and sounds. Hence, people with hearing loss could more easily participate in social life again – even in acoustically demanding situations and environments. “The AI-based target speaker extraction technology provides such selective auditory attention ability, and the neuro-steered mechanism lowers the cognitive efforts during listening,” says Li. These efforts are often challenging for people with hearing loss. They understand individual words or snippets of conversation and try to put them into context without being sure whether they are right. Conversations become laborious, and spontaneous exchange is difficult. The research results could thus help people with hearing loss not only hear more clearly but also focus more easily on conversations.

Suitable Research Environment at the University of Bremen

Li’s research field, machine listening, combines elements of computer science, cognitive science, neuroinformatics, and engineering. In the Cognitive Systems Lab (CSL) at the University of Bremen and the high-profile area “Minds, Media, Machines,” he found a suitable research environment for the project. CSL Director and Computer Science Professor Tanja Schultz emphasizes: “Our joint research in the area of selective auditory attention benefits from the complementary expertise of his team in speaker extraction and my team in interpreting brain activity using EEG.” In the first project phase, from 2019 to 2025, the teams produced more than 40 joint scientific publications on the topic, which have already been cited over 700 times.

Thanks to his excellent reputation, Li has increased the international visibility of the CSL and the University of Bremen: “He is very well connected and thus provides access to an international network with great potential for collaboration in Asia and the USA – a valuable gateway for strategic partnerships,” says Schultz. She and Li have known each other for 20 years. They have been board members of the International Speech Communication Association (ISCA) for years, even serving as president of this world’s largest association for speech processing research. “So we had a very good basis for this intercontinental cooperation,” she says.

Pioneering Research Topic

The results also offer numerous points of connection to the Hearing4All research focus at the University of Oldenburg, with which the University of Bremen is linked through the Northwest Alliance. As part of the U Bremen Excellence Chair Program, Li was recruited as a co-applicant for the Graduate School “Hearable-Centered Assistance: From Sensor to Participation,” which both universities successfully secured from the German Research Foundation (DFG) in 2024. Within this graduate school, doctoral researchers are developing and investigating a new generation of networked technical assistance systems worn on the ear – so-called hearables – that can, for example, measure biosignals in everyday life, from which health status or critical situations can be inferred – similar to a smartwatch. In another DFG project called activeEars, approved this summer, Li and Schultz and their teams now aim to improve EEG-based voice extraction when the speaker is in motion, such as a server in a restaurant. Schultz is certain that her CSL team will continue to explore selective auditory attention in the future: “It is pioneering and extremely attractive for aspiring young researchers.”

Further Information

U Bremen Excellence Chairs