Generative Data Augmentation Challenge: Synthesis of Room Acoustics for Speaker Distance Estimation

TL;DR

The paper addresses the limited diversity of room impulse response (RIR) data for training spatial audio systems, particularly speaker distance estimation (SDE). It proposes a two-task generative data augmentation pipeline that synthesizes RIRs from sparse room information and uses them to fine-tune a fixed-architecture SDE model, evaluated via direct RIR generation (Task 1) and downstream SDE usefulness (Task 2). The study provides a large, partly simulated dataset (Treble wave-based and GWA hybrid RIRs) plus a measured-control room, defines rigorous evaluation metrics (e.g., $T20$, $EDF$, $DRR$) and baselines, and reports that oracle SDE models trained on challenge data outperform the C4DM-trained baseline, suggesting generated RIRs can meaningfully boost SDE accuracy. This work demonstrates the viability of generative RIRs for improving generalization in spatial audio applications, with practical impact for AR/VR and related technologies.

Abstract

This paper describes the synthesis of the room acoustics challenge as a part of the generative data augmentation workshop at ICASSP 2025. The challenge defines a unique generative task that is designed to improve the quantity and diversity of the room impulse responses dataset so that it can be used for spatially sensitive downstream tasks: speaker distance estimation. The challenge identifies the technical difficulty in measuring or simulating many rooms' acoustic characteristics precisely. As a solution, it proposes generative data augmentation as an alternative that can potentially be used to improve various downstream tasks. The challenge website, dataset, and evaluation code are available at https://sites.google.com/view/genda2025.

Figures (2)

• Figure 1: Overview of the generative data augmentation pipeline.
• Figure 2: Images of the Treble simulated rooms.