How to (virtually) train your speaker localizer

TL;DR

This paper demonstrates that enhancing the realism of simulated training data, by extending the image source method to include more accurate acoustic responses, significantly improves the performance of speaker localization systems in real environments.

Contribution

The paper introduces a method to extend the image source method for more realistic acoustic simulations, boosting real-world speaker localization accuracy.

Findings

Increased training set realism improves localization accuracy.

Extensions to the ISM lead to consistent performance gains.

Realism layers each contribute positively to system performance.

Abstract

Learning-based methods have become ubiquitous in speaker localization. Existing systems rely on simulated training sets for the lack of sufficiently large, diverse and annotated real datasets. Most room acoustics simulators used for this purpose rely on the image source method (ISM) because of its computational efficiency. This paper argues that carefully extending the ISM to incorporate more realistic surface, source and microphone responses into training sets can significantly boost the real-world performance of speaker localization systems. It is shown that increasing the training-set realism of a state-of-the-art direction-of-arrival estimator yields consistent improvements across three different real test sets featuring human speakers in a variety of rooms and various microphone arrays. An ablation study further reveals that every added layer of realism contributes positively to these improvements.