AI headphones cut through the noise: select who and what you want to listen to

An AI system by the University of Washington lets a user single out a voice they want to listen to.

A “target speech hearing” app installed on the headphones picks up a speaker's voice after a user looks at them for three to five seconds. Once the speaker is “enrolled,” the app cancels all environmental noise. The listener hears the speaker even when they’re moving and no longer facing the speaker.

While noise-canceling earbuds and headsets are nothing new, choosing and singling out a specific speaker is not a characteristic they share.

“Our design leverages these advances in noise-canceling hearables to reduce the amplitude of all external sounds and noise and then introduce back the target speaker through the headset with end-to-end delays of less than 20 ms,” the research paper reads.

To build the prototype, researchers used a Sony WH-1000XM4 noise-canceling headset and a pair of Sonic Presence SP15C microphones.

This is an important contribution to the “overarching vision of intelligent hearables,” the paper’s authors claim. In the future, they hope that headsets and earbuds will enrich human auditory perception. In other words, a person should be able to select and hear the sounds they want in real-time.