Natural voice isolation: Scientists reveal how the brain isolates voices in noisy environments

For decades, neuroscientists have studied the "cocktail party problem," which is the human brain's remarkable ability to intentionally tune into a single voice amid a cacophony of background noise. While it has long been understood that the brain achieves this internal focus by amplifying the activity of neurons tuned to specific audio traits, researchers previously lacked a working computational model to prove this mechanism was sufficient for real-world scenarios.

Recently, a team of researchers from the Massachusetts Institute of Technology successfully developed an artificial neural network that mirrors this human auditory ability. Published in the journal Nature Human Behavior, the study reveals that the brain utilizes a strategy known scientifically as multiplicative feature gains. In simple terms, the brain acts like a highly specific volume dial. When a person listens to a target voice, their brain amplifies neural signals associated with that voice's unique characteristics, such as its pitch, while simultaneously turning down the volume on competing sounds.

To test this, the MIT team fed their artificial model a short audio cue of a specific voice, followed by a noisy mixture of overlapping speakers. The model successfully boosted the target voice to the forefront, matching human performance across diverse conditions. It even replicated common human listening errors, such as struggling to separate two distinct voices that share similar pitches.

None of our models has had the ability that humans have, to be cued to a particular object or a particular sound and then to base their response on that object or that sound. That’s been a real limitation. — Josh H. McDermott, corresponding author of the paper on this study.

The model also allowed researchers to rapidly test how spatial location affects listening. The system predicted that distinguishing between voices is significantly easier when the speakers are separated horizontally rather than vertically — a phenomenon the team subsequently confirmed in human trials. The researchers hope this model will pave the way for advanced cochlear implants that can help individuals focus their attention more effectively in chaotic environments.