Probing Spatial Structure in Pretrained Audio Representations
arXiv:2606.05544
The paper introduces SARL, a benchmark for evaluating how pretrained audio models encode spatial information such as source direction and room acoustics, and analyzes the strengths and biases of several existing encoders.
Abstract
Pretrained spatial audio encoders are increasingly used as general-purpose representations for perceptual tasks, yet their spatial encoding capabilities remain poorly understood. We introduce the Spatial Audio Representation Learning (SARL) benchmark, a controlled framework for evaluating spatial information in pretrained audio models. SARL probes source-level factors (azimuth, elevation, distance, class) and room-level factors (RT60, volume, shape). Experiments across diverse encoders reveal three patterns: input configuration and training paradigm shape spatial encoding; source factors are consistently easier to decode than room factors; and sensitivity analysis under controlled perturbations shows heterogeneous responses to source and room variation. These results reveal systematic biases in current pretrained audio representations. SARL is released as an open-source benchmark for reproducible evaluation of spatial audio representations.
Accepted to Interspeech 2026