Reconstruction of reverberant sound fields over large spatial domains.

Characterising acoustic fields in rooms is challenging due to the complexity of data acquisition. Sound field reconstruction methods aim at predicting the acoustic quantities at positions where no data are available, incorporating generalisable physical priors of the sound in a room. This study introduces a model that exploits the general time structure of the room impulse response, where a wave-based expansion addresses the direct sound and early reflections, localising their apparent origin, and kernel methods are applied to the late part. This late energy is considered to follow a sinc-like spatial correlation, in accordance with the random wave field theory. Synthesised pressure points, which follow the observed statistics of the sound field, are introduced to enable extrapolation over large distances. The model is evaluated experimentally in a lecture room and an auditorium, demonstrating a successful reconstruction of the sound field across a 5 m aperture using three microphone arrays of only 4.2 cm radius each. These results indicate that the proposed methodology enables volumetric extrapolation over several orders of magnitude, which is significant in the context of navigable sound field reproduction, "6-degrees of freedom" spatial audio and sound field analysis in rooms.