Researchers have introduced OpenWhistle, a large-scale dataset and benchmark designed to advance the study of bottlenose dolphin vocalizations. This dataset, comprising approximately 180,000 whistles recorded over five years, is the most extensive publicly available collection of dolphin sounds. It includes expert-annotated whistles and evaluation protocols for tasks such as whistle-type detection and classification. A Wav2Vec2.0 model adapted for dolphin acoustics was pre-trained on OpenWhistle, demonstrating superior performance compared to existing bioacoustic models and highlighting the potential for self-supervised learning in understanding complex animal communication. AI
IMPACT Enables new research into animal communication using advanced AI models.
RANK_REASON The cluster describes a new dataset and benchmark released via arXiv, which is a form of research output. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →