{"paper":{"title":"FSD50K-Solo: Automated Curation of Single-Source Sound Events","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A framework using diffusion-generated mixtures and a pre-trained classifier automatically filters multi-source samples from FSD50K to produce the single-source subset FSD50K-Solo.","cross_cats":[],"primary_cat":"eess.AS","authors_text":"Bryce Irvin, Li-Chia Yang, Marko Stamenovic, Ningyuan Yang, Shuo Zhang, Sile Yin, Xiao Quan","submitted_at":"2026-05-13T16:04:12Z","abstract_excerpt":"High-quality training datasets are essential for the performance of neural networks. However, the audio domain still lacks a large-scale, strongly-labeled, and single-source sound event dataset. The FSD50K dataset, despite being relatively large and open, contains a considerable fraction of multi-source samples where background interference or overlapping events could limit the usefulness of the data. To address this challenge, we introduce a data curation framework designed for large-scale open audio corpora. Our approach leverages a generative diffusion model to synthesize clean single-class"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments show that our framework achieves strong performance on a human expert-curated test set.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The pre-trained audio encoder and discriminative classifier can reliably distinguish single-source from multi-source samples when trained on diffusion-generated mixtures.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"The authors present a scalable curation method that combines diffusion-based mixture synthesis with a discriminative classifier to automatically extract single-source sound events from FSD50K and release the cleaned FSD50K-Solo subset.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A framework using diffusion-generated mixtures and a pre-trained classifier automatically filters multi-source samples from FSD50K to produce the single-source subset FSD50K-Solo.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"8ae00d3250cc2cdfa2c47dd101c00aea3cacbf99b96ca335b5564b9f24d5495a"},"source":{"id":"2605.13931","kind":"arxiv","version":1},"verdict":{"id":"180276c2-fb26-4424-b9d1-9ec627472708","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:46:22.824941Z","strongest_claim":"Experiments show that our framework achieves strong performance on a human expert-curated test set.","one_line_summary":"The authors present a scalable curation method that combines diffusion-based mixture synthesis with a discriminative classifier to automatically extract single-source sound events from FSD50K and release the cleaned FSD50K-Solo subset.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The pre-trained audio encoder and discriminative classifier can reliably distinguish single-source from multi-source samples when trained on diffusion-generated mixtures.","pith_extraction_headline":"A framework using diffusion-generated mixtures and a pre-trained classifier automatically filters multi-source samples from FSD50K to produce the single-source subset FSD50K-Solo."},"references":{"count":27,"sample":[{"doi":"","year":2025,"title":"Pseldnets: Pre-trained neural networks on a large-scale synthetic dataset for sound event localization and detection,","work_id":"1affe7d6-f32d-4240-9253-4fd46480c2d8","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Hissnet: Sound event detection and speaker identification via hierarchical prototypical networks for low-resource headphones,","work_id":"7a97c1e6-1343-4956-9881-a7b76b6782e1","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Conette: An efficient audio captioning system leveraging multiple datasets with task embedding,","work_id":"a2c839e0-8527-4abb-9567-fb54904ac4cc","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Wavcaps: A chatgpt-assisted weakly-labelled au- dio captioning dataset for audio-language multimodal research,","work_id":"b5885d8e-682b-4bfc-a808-7f3fc0ac45af","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Real-time target sound extraction,","work_id":"b0a24a79-bf89-4923-9df8-a09d28e12305","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":27,"snapshot_sha256":"d9a450525eb8a1a5a61a289d17938c10120e30f449e9bc06d6f8ce5cbe532b94","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"edf6585e7b40736339054cc53a38f7b07d5292cd561728f15afe80f6f6868824"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}