Reproduce “Spatial Scaper: A Library to Simulate and Augment Soundscapes for Sound Event Localization and Detection in Realistic Rooms” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Iran R. Roman, Christopher Ick, Sivan Ding, Adrian S. Roman, Brian McFee, Juan P. Bello
SpatialScaper is an open-source Python library that simulates and augments soundscapes for Sound Event Localization and Detection (SELD) by emulating diverse virtual acoustic environments and supporting dynamic sound sources. The library's case studies show that increased acoustic diversity in training data measurably improves SELD model localization accuracy, and channel swapping significantly enhances overall model robustness.
Key insights
- Increasing the acoustic diversity within SELD training data, particularly by adding varied room characteristics from both real and virtual RIRs, leads to significant and progressive improvements in localization error (LE).
- Channel swapping, a spatial data augmentation technique, proves highly effective in enhancing SELD model robustness, leading to substantial improvements across most SELD metrics, including Error Rate, F-score, and Localization Error.
- While virtual room simulations offer significant flexibility, models trained exclusively on data from purely virtual RIRs perform substantially worse than those trained on real RIRs, indicating that current geometric acoustic models may not fully capture the complexities of real room acoustics for generalization.
- Pitch-shifting as an augmentation technique shows minimal to no benefit for SELD model performance, suggesting its limited utility compared to spatial or spectral augmentations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.