Noise-canceling headphones are excellent at blocking sound but struggle to selectively allow certain noises. The latest Apple AirPods Pro adjust sound levels automatically but offer limited user control over who to listen to.
A University of Washington team has developed “Target Speech Hearing,” an AI system that lets headphone users focus on a specific speaker. By looking at the person speaking for three to five seconds, the system “enrolls” them, canceling other sounds and amplifying the speaker’s voice in real-time, even as the listener moves.
Presented on May 14 at the ACM CHI Conference, the system’s code is available for further development but isn’t commercially available. Shyam Gollakota, a UW professor, explained that this AI modifies auditory perception based on user preferences, allowing clear hearing of a single speaker in noisy environments.
Using off-the-shelf headphones with microphones, users tap a button while facing the speaker. The sound is sent to an onboard computer, where machine learning software identifies the speaker’s voice. The system improves its focus as the speaker continues talking, providing more data.