Researchers have developed a system that reads a listener’s brain activity to pick out which voice to amplify in a noisy environment — a long-standing challenge for hearing-assistive technology. Reported in Nature Neuroscience, the approach uses patterns of neural activity that reveal which speaker a person is attending to and uses that information to boost the chosen voice and suppress other sounds.
The work builds on a discovery from 2012 showing that, in people with typical hearing, the auditory cortex produces a distinct pattern of brain waves that tracks the speech stream a listener is focusing on. That neural pattern acts like a signature of attention: by monitoring it, researchers can tell which speaker the listener wants to hear.
The new study was led from Columbia University by Nima Mesgarani, with major contributions from graduate student Vishal Choudhari. To test their idea the team used patients who already had electrodes placed on the surface of their brains for epilepsy treatment. Those electrodes allowed direct recording from the auditory cortex while the researchers simulated a “cocktail party” at the bedside — two different conversations played simultaneously from separate speakers.
At first, both conversations were at the same level, making comprehension difficult. The research system decoded the brain signals in real time to determine which conversation each participant was attending to, then automatically increased the volume of that conversation and decreased the other. The system identified the intended speaker correctly up to about 90% of the time. When it was active, participants’ comprehension improved and they reported less listening effort.
All four people tested had typical hearing, so the results are a proof of concept rather than a ready-made consumer product. Josh McDermott, who leads the Laboratory for Computational Audition at MIT and was not involved in the study, notes that whether the approach will work as well for people with hearing loss is still an open question. Hearing loss can change the strength or clarity of neural signals, which might make decoding harder.
Still, the idea points toward a new class of “brain-controlled” hearing devices — hearing aids, cochlear implants and other assistive listening systems that use a listener’s neural signature of attention to decide which sound to enhance. Existing hearing aids do fairly well at suppressing steady background noise but struggle to choose between competing voices because they lack information about which speaker a user is trying to follow.
There are other possible routes to the same goal. For example, artificial intelligence might learn a particular user’s behavior and preferences and predict which voice they want to hear without direct neural signals. But a brain-based signal would give a direct, moment-by-moment readout of attention, which could be especially powerful in complex listening scenes.
Wider adoption faces technical and practical hurdles. The current study used intracranial electrodes, which are invasive and not feasible for routine use. Future work will need noninvasive ways to capture the relevant brain signals reliably, tests on people with different degrees of hearing loss, and engineering to fit decoding systems into wearable devices.
The need is large: many older adults live with disabling hearing loss, and the ability to understand conversation in noisy places remains one of the most common and frustrating limitations of today’s hearing technology. By tapping the brain’s own attention signals, researchers hope to create smarter hearing systems that let users cut through the noise and hear the person they want to listen to.