Room geometry estimation

In this paper, we target the problem of room geometry estimation from acoustic events in shoebox rooms. We consider scenarios with stationary sound sources playing known sounds, and scenarios where a moving sound source plays an unknown sound. Our main technical contribution is two solvers to find walls given three, respectively two, pairs of distances between microphones and speakers corresponding to echoes in a wall. The second solver uses an orthogonality constraint on the wall to reduce the measurements needed. These solvers are used in a hypothesis and test framework based on RANSAC, in order to robustly handle outliers and noise in the data. We also present a pseudo-synthetic dataset of sounds simulated in environments based on real rooms in the ScanNet++ dataset.

This work was partially supported by the strategic research projects ELLIIT and Wallenberg AI, Autonomous Systems and Software Program (WASP) funded by the Knut and Alice Wallenberg Foundation.