Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Physics-Informed Neural Network-Based Approach for the Spatial Upsampling of Spherical Microphone Arrays

Federico Miotello, Ferdinando Terminiello, Mirco Pezzoli, Alberto Bernardini, Fabio Antonacci, Augusto Sarti

TL;DR

This work tackles the problem of achieving high spatial resolution with a spherical microphone array using few capsules by employing a physics-informed neural network (PINN) with Rowdy activations to upsample the acoustic field. The method uses two parallel SIREN networks to model the real and imaginary parts of the complex pressure field and enforces the Helmholtz equation as a PDE regularizer, enabling accurate reconstruction of high-order spherical-harmonics content from limited measurements. Experiments on measured RIR data from an Eigenmike EM32 show that the proposed approach consistently outperforms a state-of-the-art signal-processing upsampling method (SARITA) across multiple sensing configurations, with improvements up to several decibels in NMSE. The work demonstrates the practical viability of physics-guided neural representations for space-time audio and suggests avenues for incorporating additional physical constraints to further bolster performance in challenging environments.

Abstract

Spherical microphone arrays are convenient tools for capturing the spatial characteristics of a sound field. However, achieving superior spatial resolution requires arrays with numerous capsules, consequently leading to expensive devices. To address this issue, we present a method for spatially upsampling spherical microphone arrays with a limited number of capsules. Our approach exploits a physics-informed neural network with Rowdy activation functions, leveraging physical constraints to provide high-order microphone array signals, starting from low-order devices. Results show that, within its domain of application, our approach outperforms a state of the art method based on signal processing for spherical microphone arrays upsampling.

A Physics-Informed Neural Network-Based Approach for the Spatial Upsampling of Spherical Microphone Arrays

TL;DR

This work tackles the problem of achieving high spatial resolution with a spherical microphone array using few capsules by employing a physics-informed neural network (PINN) with Rowdy activations to upsample the acoustic field. The method uses two parallel SIREN networks to model the real and imaginary parts of the complex pressure field and enforces the Helmholtz equation as a PDE regularizer, enabling accurate reconstruction of high-order spherical-harmonics content from limited measurements. Experiments on measured RIR data from an Eigenmike EM32 show that the proposed approach consistently outperforms a state-of-the-art signal-processing upsampling method (SARITA) across multiple sensing configurations, with improvements up to several decibels in NMSE. The work demonstrates the practical viability of physics-guided neural representations for space-time audio and suggests avenues for incorporating additional physical constraints to further bolster performance in challenging environments.

Abstract

Spherical microphone arrays are convenient tools for capturing the spatial characteristics of a sound field. However, achieving superior spatial resolution requires arrays with numerous capsules, consequently leading to expensive devices. To address this issue, we present a method for spatially upsampling spherical microphone arrays with a limited number of capsules. Our approach exploits a physics-informed neural network with Rowdy activation functions, leveraging physical constraints to provide high-order microphone array signals, starting from low-order devices. Results show that, within its domain of application, our approach outperforms a state of the art method based on signal processing for spherical microphone arrays upsampling.
Paper Structure (9 sections, 11 equations, 1 figure, 1 table)

This paper contains 9 sections, 11 equations, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Problem formulation
  3. Data model
  4. Spatial upsampling of sma
  5. Method
  6. Evaluation
  7. Experiments setup
  8. Results and discussion
  9. Conclusion

Figures (1)

  • Figure 1: Comparison between ground truth RIR and RIRs reconstructed using (a) SARITA and (b) the proposed method, starting from 9 available sma channels.