Imagine standing in a crowded room where voices blur into a single noisy hum. Most people can focus on one speaker and their brain boosts that voice while suppressing others. That everyday ability, known as the cocktail party problem, is much harder for people who rely on hearing aids.
A study published in Nature Neuroscience describes a system that decodes brain activity to determine which voice a listener is attending to, then amplifies that voice for the listener. Researchers led by Nima Mesgarani at Columbia University call it a brain-controlled hearing system. The approach could eventually improve hearing aids, assistive listening devices and cochlear implants.
The new work builds on a 2012 discovery by Mesgarani and Dr. Eddie Chang showing that the auditory cortex produces a distinct pattern of brain waves that track the particular speech stream a listener is focusing on. That neural pattern acts like a signature: by reading it, researchers can tell which speaker a person is attending to.
To test the idea, the team worked with four hospital patients who already had electrodes implanted in their brains for epilepsy treatment. The electrodes let researchers record signals directly from the auditory cortex. At the bedside they simulated a cocktail party by playing two different conversations from separate speakers. When both conversations played at equal volume, the participants had trouble understanding either one.
The team then switched on a system that monitored the participants’ brain waves and used the neural signature to decide which conversation to amplify. If the recorded activity indicated the listener was attending to conversation one, the system raised that speaker’s volume and reduced the other. The brain-decoding algorithm identified the intended conversation with up to 90 percent accuracy. When the system was active, participants understood more and reported less listening effort.
Experts caution the results are preliminary. The experiment involved only four people with typical hearing who happened to have clinical electrodes implanted for unrelated reasons. Josh McDermott of MIT, who was not part of the study, notes that people with hearing loss may produce weaker neural signals, so it remains an open question how well the method will translate to users of hearing aids.
Still, current hearing aids and implants struggle with competing voices. They include algorithms that reduce steady background noise, but they lack a reliable way to identify which voice the listener wants to hear. Brain-controlled decoding offers one path to solve that problem by directly reading the listener’s focus. Another possible route is to use artificial intelligence to learn a person’s typical behavior and predict the likely target voice without neural input.
The demand for better solutions is growing: more than half of people 75 and older live with disabling hearing loss, and age-related hearing decline is common. The new study demonstrates a promising proof of concept that the brain’s attention signals can be used to steer hearing technology. But more research is needed to test noninvasive sensors, larger and more diverse groups, and whether the approach can be adapted for people with hearing impairment before it becomes a practical option for everyday devices.