Reproduce “Zero- and Few-shot Sound Event Localization and Detection” 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
Kazuki Shimada, Kengo Uchida, Yuichiro Koyama, Takashi Shibuya, Shusuke Takahashi, Yuki Mitsufuji, Tatsuya Kawahara
The authors present embed-ACCDOA for zero- and few-shot sound event localization and detection (SELD). It predicts event tracks pairing CLAP-aligned embeddings with directions, allowing text or example recordings to label sources in overlapping audio. On STARSS23 with 10-shot support, it improves location-dependent error rate and F-score over CLAP plus DOAE, but has lower localization recall and similar aggregate error; it remains behind the full-data baseline.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.