Deep Learning-Assisted Two-Cavity Method for Estimating Sound Propagation Characteristics in Porous Media

This study presents a method for estimating sound propagation characteristics in highly porous materials based on a single measurement with a two-microphone impedance tube using the standardized transfer function method. The proposed method adopts the principle of the two-cavity method but substitutes the second experimental measurement with a deep neural network. The selected network architecture enables predictions of the sound propagation characteristics at the measurement’s frequency resolution in a single forward pass. Trained on purely synthetic data, this hybrid two-cavity method provides accurate estimates of the sound propagation characteristics, with relative errors below 10% for most simulated materials and experimentally tested melamine foam. This study marks the first step towards a generalized deep learning-based assistant for acoustic material characterization.