OpenWhistle: largest open dolphin-vocalization dataset (180k whistles, 114h) is a NeurIPS Spotlight
emaros96 · x · 2026-10-22
OpenWhistle, a NeurIPS 2026 Datasets & Evaluations Spotlight (top 1.3%), is the largest publicly available dolphin vocalization dataset: 180,000 whistles (114 hours) recorded over five years from a stable pod of five bottlenose dolphins, with 8,354 expert-annotated whistles and a full detection/segmentation/classification pipeline. Motivated by the lack of long-term, single-species depth in bioacoustics corpora, the team pretrained an adapted Wav2Vec2.0 on the corpus, beating general bioacoustic models like AVES and BioLingual on both whistle-type classification (81.1%) and detection (75.6 mAP). Data, code, and paper are all open.
More from Research
- CONCERT: Generative Modeling of Spatially Resolved Perturbation Responses — anshulkundaje · 2026-10-22
- ToM2Vec trained via self-play with Qwen3-8B transfers to GPT-5.4 teams, adding 17.3 points — EliasEskin · 2026-10-22
- ToM2Vec: continuous theory-of-mind vectors help LLM agents coordinate with unseen partners — EliasEskin · 2026-10-22
- Hands-on: building protein design workflows with RFD4-Proteina on Hugging Face Spaces — AllThingsApx · 2026-10-22
- edgePython v0.5.0 released: 4-6x faster than edgeR, camera over 100x speedup — lpachter · 2026-10-22
- 'Agentic Inequality' paper published in Nature Machine Intelligence: who captures AI's surplus agency — IasonGabriel · 2026-10-22