{"ID":2868605,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2509.15599","arxiv_id":"2509.15599","title":"MAGENTA: Magnitude and Geometry-ENhanced Training Approach for Robust Long-Tailed Sound Event Localization and Detection","abstract":"Deep learning-based Sound Event Localization and Detection (SELD) systems degrade significantly on real-world, long-tailed datasets. Standard regression losses bias learning toward frequent classes, causing rare events to be systematically under-recognized. To address this challenge, we introduce MAGENTA (Magnitude And Geometry-ENhanced Training Approach), a unified loss function that counteracts this bias within a physically interpretable vector space. MAGENTA geometrically decomposes the regression error into radial and angular components, enabling targeted, rarity-aware penalties and strengthened directional modeling. Empirically, MAGENTA substantially improves SELD performance on imbalanced real-world data, providing a principled foundation for a new class of geometry-aware SELD objectives. Code is available at: https://github.com/itsjunwei/MAGENTA_ICASSP","short_abstract":"Deep learning-based Sound Event Localization and Detection (SELD) systems degrade significantly on real-world, long-tailed datasets. Standard regression losses bias learning toward frequent classes, causing rare events to be systematically under-recognized. To address this challenge, we introduce MAGENTA (Magnitude And...","url_abs":"https://arxiv.org/abs/2509.15599","url_pdf":"https://arxiv.org/pdf/2509.15599v1","authors":"[\"Jun-Wei Yeow\",\"Ee-Leng Tan\",\"Santi Peksi\",\"Woon-Seng Gan\"]","published":"2025-09-19T05:00:20Z","proceeding":"eess.AS","tasks":"[\"eess.AS\",\"cs.SD\"]","methods":"[]","has_code":false,"code_links":[{"ID":609605,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_id":2868605,"paper_url":"https://arxiv.org/abs/2509.15599","paper_title":"MAGENTA: Magnitude and Geometry-ENhanced Training Approach for Robust Long-Tailed Sound Event Localization and Detection","repo_url":"https://github.com/itsjunwei/MAGENTA_ICASSP","is_official":false,"mentioned_in_paper":false,"mentioned_in_github":true,"github_stars":0}]}