Seeing with Sound

Research led by Dr. Iran R. Roman within the School of Electronic Engineering and Computer Science gives machines a new superpower: the ability to visualise sound. While humans rely heavily on sight, the acoustic world contains a wealth of invisible information. This research bridges the gap between audio signal processing, computer vision, and artificial intelligence to create high-definition "acoustic images."

Typically, visualising sound requires expensive, heavy equipment with dozens of microphones. However, we leverage a breakthrough AI technique called Latent Acoustic Mapping (LAM) that can "super-resolve" audio. We are now working on allowing lightweight wearable devices (specifically Meta’s NEW Aria smart glasses) to generate 3D maps of sound sources in real-time. This technology is powered by AudibleLight, a pioneering open-source library developed at QMUL that simulates complex audiovisual environments.

This transformative work, developed in collaboration with Meta Reality Labs, has direct applications for "Tomorrow's World." By enabling wearables to accurately locate and visualise sound sources (even those outside a user’s field of view) we are paving the way for safer Augmented Reality (AR) experiences and smarter hearing assistance technologies that can focus on specific speakers and sounds in noisy environments.

• More information: Centre for Multimodal AI
• Contact: Dr. Iran Roman