I-Corps: Translation potential of accurate acoustic digital twins for complex real-world environments

This I-Corps project is based on the development of a technology that can reproduce the sound of physical spaces from a small number of acoustic measurements. Every space — a cathedral, a concert hall, a classroom, a restaurant, a historic site — has a distinctive acoustic signature that shapes how music, speech, and everyday sounds are experienced within it. Currently, capturing that signature requires either exhaustive on-site measurement campaigns that are slow and expensive, or synthetic approximations that sound generic and unconvincing. This technology combines efficient acoustic capture with advanced computational reconstruction, producing high fidelity representations of real spaces at a fraction of the cost and time required by current methods. Applications include immersive media production, architectural and hospitality design, hearing-device development, cultural heritage preservation, and machine listening. This I-Corps project utilizes experiential learning coupled with first-hand investigation of the industry ecosystem to assess the translation potential of a Bayesian machine learning framework for reconstructing acoustic fields in real physical spaces from sparse impulse-response measurements. This technology combines Gaussian-process priors trained across a curated corpus of measured spaces with physically motivated wave-equation constraints, yielding perceptually convincing and physically consistent auralizations at arbitrary listener positions, together with calibrated estimates of reconstruction uncertainty. This methodology departs from purely data-driven neural approaches, which are opaque and often physically inconsistent, and from purely geometric acoustics models, which fail at low frequencies and in complex geometries. The framework is complemented by a robotic capture platform that standardizes and accelerates on-site measurement, dramatically reducing the number of measurements required to characterize a new space. Users may benefit from the ability to produce faithful acoustic representations of real environments with a fraction of the measurement effort, expertise, and cost previously required, and from calibrated uncertainty estimates that quantify confidence in reconstructions at unmeasured listening positions. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.