Research map / Label-efficient and robust learning / Self-supervised and contrastive
AVEX: What Matters for Animal Vocalization Encoding
Marius Miron, David Robinson, Milad Alizadeh, Ellen Gilsenan-McMahon, Gagan Narula, Emmanuel Chemla, Maddie Cusimano, Felix Effenberger, Masato Hagiwara, Benjamin Hoffman, Sara Keen, Diane Kim and 5 more
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 AudioSet general audio to supervised post-training data (ablation) | EffNetB0 · BEANS Detection / Individual ID / Vocal Repertoire (aggregate) | probe / retrieval: Individual ID probe 0.457 -> 0.531; Vocal R-auc 0.806 -> 0.830 | Improved |
| post-train on AudioSet only instead of bioacoustic data (ablation) | EffNetB0 · BEANS Classification (probe) | accuracy: 0.786 -> 0.651 | Hurt |
| add AudioSet to SSL pre-training data (ablation) | EAT · BirdSet (probe) | mAP: 0.143 -> 0.166 | Improved |
| add supervised post-training on bioacoustic data to SSL backbone (ablation) | BEATs / EAT · BEANS Classification (probe) | accuracy: 0.774 -> 0.840 (BEATs -> sl-BEATs-bio) | Improved |
| add AudioSet to the post-training mix (ablation) | BEATs · BEANS Classification (probe) | accuracy: 0.840 -> 0.832 (BEATs); Detection 0.390 -> 0.408 | Mixed |
| add AudioSet to the post-training mix (ablation) | EAT · Individual ID (probe) | accuracy: 0.495 -> 0.456 | Hurt |
| add soundscape data to the supervised training mix (ablation) | EffNetB0 · BEANS / BirdSet (appendix heatmaps) | performance change | No clear effect |
| use SSL pre-training instead of supervised encoders for focal-to-soundscape shift | SSL vs supervised encoders · BEANS Classification -> Detection retrieval | ROC AUC drop: drop 0.01 (SSL) vs 0.09 (supervised) | Improved |
| AVEX best post-trained model vs Perch/BirdNet/NatureBEATs | sl-BEATs-all · BEANS Detection (probe) | mAP: 0.392 -> 0.408 | Improved |
Closest papers by meaning
- NatureLM-audio: an Audio-Language Foundation Model for Bioacoustics (ICLR 2025 · similarity 0.93)
- OpenWhistle: A Large-Scale Longitudinal Dataset and Benchmark of Bottlenose Dolphin Vocalizations (NeurIPS 2026 · similarity 0.90)
- BirdSet: A Large-Scale Dataset for Audio Classification in Avian Bioacoustics (ICLR 2025 · similarity 0.88)
- Vocal Call Locator Benchmark (VCL) for localizing rodent vocalizations from multi-channel audio (NeurIPS 2024 · similarity 0.88)
- The iNaturalist Sounds Dataset (NeurIPS 2024 · similarity 0.87)
- Towards A Translative Model of Sperm Whale Vocalization (NeurIPS 2025 · similarity 0.85)
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.
