ARPA-H HEARING: a brain-driven hearing interface, attentional focus, and age-related hearing loss

ARPA-H launches HEARING: a brain-driven hearing interface that must detect which voice a person is attending to, suppress noise, and help with age-related hearing loss

On 7 May, the U.S. agency ARPA-H opened a competition for a system for people with mild, moderate, and severe hearing loss. The device is meant to work with the brain’s auditory cortex: tracking attentional focus, adjusting the soundscape around a person in real time, and restoring missing pieces of speech.

A conventional hearing aid amplifies sound. In a quiet room, that is often enough. In a restaurant, at a family dinner, or in a hospital corridor, it amplifies the conversation partner’s voice, the next table over, clattering dishes, and the air conditioner alike. The person hears more noise, even though what they need is a single voice.

ARPA-H frames the problem differently: hearing depends on the ear and the brain working together. The brain has to separate speech from noise, choose the source, and fill in missing fragments of sound. In the HEARING program, the agency wants to build a minimally invasive neural interface that tackles this part of the job directly in the auditory cortex.

In April, the U.S. regulator already cleared Otarmeni — gene therapy for a rare congenital deafness, where a single protein involved in sound transmission in the inner ear is defective. HEARING targets a different part of the auditory pathway: the brain that picks out the right voice in noise.

The system has three layers. The first is a cortical device that captures signals in the brain. The second is an algorithm that reshapes the auditory scene: boosting the target voice, muting background noise, and adapting sound to the situation. The third is a read/write model: it must read what the person is attending to and return missing auditory information.

In user terms, it sounds simpler. A person sits in a noisy room and turns toward the person they want to hear; the brain is already singling that voice out from the mix. The device reads that choice and makes that same voice louder and clearer. If speech drops out in the noise, the system tries to reconstruct the missing fragment.

The aging angle here is direct. According to ARPA-H, more than 70 million people in the U.S. live with hearing loss. The agency puts the associated costs, lost wages, and lost productivity at more than $200 billion a year. For older adults, it is one of the most common health problems; ARPA-H explicitly links it to social isolation, depression, cognitive decline, and dementia risk.

The dementia link needs to be handled carefully. The large ACHIEVE study in The Lancet tested hearing aids and counseling in 977 people aged 70–84. In the overall group, cognitive decline over three years did not differ from the control group. In a prespecified analysis, the intervention looked more beneficial in people with higher baseline risk. Based on these data, hearing care may help some older adults, but it is not yet proven as dementia prevention for everyone.

The technical idea behind HEARING already has a backstory. In 2025, a group in Journal of Neural Engineering described a system in which EEG recorded from the scalp decoded which of two speakers a person was listening to, while a microphone array adjusted the volume of the streams. In demonstrations, people with hearing impairment could shift attention, and the system amplified the desired voice.

That same work also showed why a consumer device is still a long way off. The decoder needs seconds, errors are frustrating for the user, some people produce noisy signals, and movement in daily life disrupts EEG far more than lab recordings do. If the algorithm gets it wrong, it amplifies the wrong voice and suppresses the one the person was trying to hear. For a hearing aid, that is catastrophic: the device gets in the way of conversation at the exact moment it is supposed to save it.

That is why ARPA-H’s choice looks logical. The agency is taking on the full stack: brain signals, microchips, machine learning, neuromodulation, clinical audiology, and the regulatory path. Applications for the program are due in summer 2026.

If HEARING works, a hearing aid will turn from a volume amplifier into an attention neuroprosthesis. It will help an older person not just hear sound, but stay in the conversation in a world with too much noise.