A neural network-augmented two-microphone impedance tube method to estimate sound propagation characterisics

Two-microphone impedance tubes are a popular method to measure the normal incidence sound absorption coefficient. However, a single measurement with this method does not offer enough information to identify the sound propagation characteristics, i.e. the characteristic impedance and the wave number of absorbent samples. Instead, more elaborate measurement techniques that require more time, more equipment and enhanced user knowledge need to be applied. This contribution presents a technique combining neural networks and the two-microphone impedance tube method to estimate the propagation characteristics of a highly porous sample from a single two-microphone impedance tube measurement. The measured surface impedance serves as input to a U-net architecture, which preserves the high frequency resolution of the data and returns an estimate of the surface impedance mimicking an air cavity attached behind the sample. Then, the two cavity method is used to calculate the propagation characteristics from the measured and the networks predicted surface impedances. The presented approach gives precise estimates of the propagation characteristics of simulated highly porous samples and for two different, physically measured melamine foam samples, while drastically reducing the required resources and expenses. In this way, it provides an efficient way to estimate the propagation characteristics of absorbent samples.