Project title: Physics-informed Neural Networks for Sound Field Reconstruction
Description: The objective of this project is to develop a methodology to predict the sound field in a room, based on a set of sparse measurements distributed about the room. We plan to explore the use of Physics Informed Neural Networks (PINNs) for the task of reconstructing the sound field, based on a set of sparse measurements. The main outcomes of this project are threefold:
- To curate a data set for the training of neural networks in reverberant environments.
- To examine the predictive potential of PINNs to interpolate and extrapolate acoustic data in fine spatial grids.
- To examine the potential of extrapolated data to improve sound field reproduction with an array of loudspeakers.
An additional side-objective is to examine the ability of the network to model the propagation of sound in free-fields, which is relevant for sound radiation and noise control.
All outcomes of the work as well as the training and validation data will be made openly available through the dataset sharing portal of DTUs. We foresee that the results will likely be also published in the scientific literature. The outcomes of the work have an impact in spatial audio technology, speech enhancement, sound field reproduction, auralization and noise control. As such, we believe that the work addressed in the project directly benefits a large portion of the Danish Sound Cluster membership.
Start: 1 September 2022
End: 30 November 2022
Funding: 162.000 DKK
Danish Sound Cluster professional focus area:
Uddannelses- og Forskningsministeriet (UFS)
Danish Agency for Higher Education and Science