Reproduce “Revisiting SSL for sound event detection: complementary fusion and adaptive post-processing” 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
Hanfang Cui, Longfei Song, Li Li, Dongxing Xu, Yanhua Long
This research enhances Sound Event Detection by integrating diverse self-supervised learning models through complementary fusion and introducing an adaptive post-processing technique, nSEBBs. The approach yields improved detection accuracy and a substantial reduction in computational runtime.
Key insights
- While a strong general-purpose SSL model like BEATs is effective alone, strategic dual-modal fusion (e.g., BEATs with WavLM) provides complementary performance gains, whereas indiscriminate full fusion can degrade performance.
- Speech-pretrained SSL models, despite poor individual generalization, can mitigate domain-specific biases and enhance overall SED performance across diverse sound events when fused with general audio SSL models.
- The adaptive nSEBBs post-processing method maintains or slightly improves detection accuracy while significantly reducing computational runtime by 93.8% compared to traditional static-parameter SEBBs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.