Research map / Label-efficient and robust learning / Recognition and long-tail
MetaPerch: Learning from metadata for bioacoustics foundation models
Mustafa Chasmai, Vincent Dumoulin, Jenny Hamer
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add location, season and background-species metadata prediction losses | BioBaseline · BirdSet (trained head) mean | ROC-AUC: 0.891 -> 0.906 | Improved |
| add metadata auxiliary losses | BioBaseline · BEANS classification mean | accuracy: 0.854 -> 0.870 | Improved |
| add metadata auxiliary losses | BioBaseline · BEANS detection mean | cmAP: 0.506 -> 0.512 | Improved |
| add metadata auxiliary losses | BioBaseline · WABAD (linear probe) | ROC-AUC: 0.741 -> 0.811 | Improved |
| add metadata auxiliary losses | BioBaseline · WABAD (prototype head) | ROC-AUC: 0.928 -> 0.946 | Improved |
| add metadata auxiliary losses | BioBaseline · WABAD (1-shot) | ROC-AUC: 0.912 -> 0.917 | Improved |
| add metadata losses (underrepresented regions PER, UHH, NES) | BioBaseline · BirdSet | ROC-AUC: 0.025 average gain vs 0.006 for HSN, NBP, SSW, SNE | Improved |
| add metadata losses on desert recordings | BioBaseline · WABAD Deserts / Xeric Shrublands | ROC-AUC: 0.936 -> 0.912 | Hurt |
| add metadata losses on Hainan Gibbons | BioBaseline · BEANS Gibbons | cmAP: 0.519 -> 0.500 | Hurt |
| use tuned loss weights instead of dynamic equalization (ablation) | MetaPerch · WABAD (prototype head) | ROC-AUC: 0.905 -> 0.946 | Improved |
| use tuned loss weights instead of gradient surgery (ablation) | MetaPerch · WABAD (prototype head) | ROC-AUC: 0.944 -> 0.946 | No clear effect |
| mix metadata when present in any recording rather than iff all (ablation) | MetaPerch · WABAD (prototype head) | ROC-AUC: 0.945 -> 0.946 | No clear effect |
Closest papers by meaning
- NatureLM-audio: an Audio-Language Foundation Model for Bioacoustics (ICLR 2025 · similarity 0.88)
- The iNaturalist Sounds Dataset (NeurIPS 2024 · similarity 0.86)
- OpenWhistle: A Large-Scale Longitudinal Dataset and Benchmark of Bottlenose Dolphin Vocalizations (NeurIPS 2026 · similarity 0.86)
- Vocal Call Locator Benchmark (VCL) for localizing rodent vocalizations from multi-channel audio (NeurIPS 2024 · similarity 0.83)
- Combining Observational Data and Language for Species Range Estimation (NeurIPS 2024 · similarity 0.83)
- BirdSet: A Large-Scale Dataset for Audio Classification in Avian Bioacoustics (ICLR 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
