Posted:
Researchers at Columbia University published a study in Nature Neuroscience demonstrating a real-time system that reads EEG brain wave signals to identify which speaker a person is trying to listen to and automatically amplifies that voice while reducing others.
The system addresses what researchers call the cocktail party problem, the difficulty of isolating a single voice in a noisy multi-speaker environment. This is a known limitation of conventional hearing aids, which typically amplify all sounds within a range and do not distinguish between attended and unattended speech. The researchers used electrodes placed on the brain's auditory cortex to detect distinctive wave patterns that emerge when a person selectively attends to one voice. A machine learning model trained on these patterns then fed real-time instructions to the hearing system. The researchers report the system correctly identified which speaker the person was attending to approximately 90% of the time. Participants showed improved speech comprehension and reported less listening effort when the system was active.
Several caveats apply. The trial involved patients who already had electrodes implanted for epilepsy monitoring, meaning it was not tested on typical hearing aid users or people with hearing loss in everyday settings. Non-invasive EEG sensors, which would be required for any practical hearing aid application, deliver lower signal quality than implanted electrodes. The researchers acknowledge meaningful technical gaps remain between the current proof-of-concept and any clinical or consumer application.
Read the full NPR article here.
