Visual Hearing for Context-Aware Cognitive Audio Technology (WeHear)

Research Project, 2026 – 2029

WeHear proposes a novel Virtual World-enabled hearing-assistive technology for smart glasses, designed to generate interactable and socially intelligent reconstructions of the user’s acoustic environment. The system integrates head-mounted cameras and microphone arrays with in-ear monitors. It combines egocentric vision AI with spatial audio technology to render a navigable acoustic scene directly to the user’s ears with enhanced communication partners. WeHear develops core AI components for this technology, including egocentric vision for identifying communication contexts, speech source separation, and real-time sound field reconstruction. As a key innovation, WeHear will develop egocentric vision AI capable of identifying and tracking communication partners over time, even in the presence of occlusions. This avoids the need for users to orient their head or gaze toward relevant sound sources and represents a significant step toward context-aware communication technology that adapts the sound environment to the user’s social context during everyday activities. WeHear will develop functional demonstrators of this video-informed hearing technology. The project will create and publish a large-scale open dataset from human communication experiments with AI benchmarks to foster an open innovation ecosystem for future co-creation of cognitive hearing technologies. Together, the project aims to lay the foundation for next-generation Virtual World hearing instruments that empower hearing-impaired users to communicate effectively in complex, noisy environments.