Hearing is a remarkable feat of organic engineering. Vibrating objects in our environment produce small pressure waves that vibrate our eardrum. The inner ear transforms these movements into neural activity patterns that represent the frequency and timing of sound in intricate detail. A hierarchy of auditory brain regions then performs computations that extract meaningful auditory features, such as the melody of music or the location of a ringing phone. Activity in the auditory cortex thereby gives rise to our conscious experience of a rich acoustic world.

This process becomes much more challenging when multiple sound sources are present, as we’ve all experienced when chatting with a friend in a noisy restaurant. Their voice overlaps with those of other people, background music, and clinking cutlery, forming a single acoustic mixture at the ear. In this everyday scenario, the brain must not simply encode these acoustical details, but also disentangle your friend’s voice from competing sound sources. This is often called the “cocktail party problem”. The auditory system uses features such as spatial location and pitch to segregate different sound sources, and attending to a voice selectively strengthens its neural representation. This is thought to arise through interactions between auditory cortex and attentional brain networks, but the precise mechanisms remain poorly understood.

Young, healthy listeners perform this task almost effortlessly, yet even the best AI-based closed captioning fails (often to great comedic effect) to translate speech in noise. Clinically, individuals with even mild hearing loss often struggle to follow conversations in noisy environments. In fact, research in our lab shows that these problems are common from middle age onwards, even in people with clinically “healthy” audiograms. This suggests that the issue lies not only in the ear’s ability to detect sound, but in how the brain handles complex auditory information. These impairments can lead older listeners to avoid social situations, which may help explain why hearing loss is now recognised as the largest modifiable risk factor for dementia. 

Modern hearing aids improve hearing by boosting sound levels, but provide little help in complex listening environments where multiple voices compete. Emerging research is beginning to address these limitations by shifting focus from the ear to the brain. A promising direction comes from the labs of Mesgarani and Chang, who have shown that it is possible to decode the speaker to which a listener is attending from their auditory cortical activity. They are developing closed-loop systems that feed this brain-derived information back into the listener’s hearing aid, allowing it to selectively amplify the attended voice. This fundamental shift from passive amplification to active, brain-guided hearing could improve communication in natural settings for many of the 18 million people in the UK with hearing impairments.

However, significant challenges remain. The current systems require invasive brain recordings and slow, computationally demanding decoding, making them impractical for most patients. Translating these findings into fast, wearable, non-invasive devices will require advances in both neuroscience and engineering. A deeper understanding of how pitch and other auditory features are processed in healthy listeners will be essential for designing algorithms that benefit from the brain’s natural strategies.

By moving beyond amplification and toward intelligent, brain-informed solutions, the future of hearing research may finally restore not just hearing, but understanding.