Realistic sources, receivers and walls improve the generalisability of virtually-supervised blind acoustic parameter estimators

TL;DR

This study demonstrates that increasing the realism of simulated training data, including source, receiver, and wall responses, significantly enhances the generalization of blind acoustic parameter estimators to real-world environments.

Contribution

The paper shows that training solely on highly realistic simulated data enables neural networks to accurately estimate acoustic parameters in real environments, reducing reliance on real annotated data.

Findings

Realistic simulation improves estimation accuracy on real data.

Layered realism in training data enhances generalizability.

Simulated data can replace real measurements for training.

Abstract

Blind acoustic parameter estimation consists in inferring the acoustic properties of an environment from recordings of unknown sound sources. Recent works in this area have utilized deep neural networks trained either partially or exclusively on simulated data, due to the limited availability of real annotated measurements. In this paper, we study whether a model purely trained using a fast image-source room impulse response simulator can generalize to real data. We present an ablation study on carefully crafted simulated training sets that account for different levels of realism in source, receiver and wall responses. The extent of realism is controlled by the sampling of wall absorption coefficients and by applying measured directivity patterns to microphones and sources. A state-of-the-art model trained on these datasets is evaluated on the task of jointly estimating the room's volume, total surface area, and octave-band reverberation times from multiple, multichannel speech recordings. Results reveal that every added layer of simulation realism at train time significantly improves the estimation of all quantities on real signals.