AnimalCLAP Dataset
收藏资源简介:
AnimalCLAP数据集由大阪大学和东京大学等机构联合构建,是一个涵盖6823个物种、4225小时动物声音记录的生物声学数据集,每条数据均标注了22种生态特征(如食性、活动模式等)。数据来源于iNaturalist和Xeno-canto平台,经过严格的物种筛选和GPT-5辅助标注验证,采用Creative Commons许可协议。该数据集通过整合生物分类学层次结构,旨在提升未见物种的识别能力,并支持从声学信号推断生态特征的跨学科研究,为生物多样性监测和计算生态学提供了重要资源。
AnimalCLAP is a bioacoustic dataset jointly constructed by institutions including Osaka University, The University of Tokyo, and other research organizations. It encompasses 6,823 species and 4,225 hours of animal sound recordings, with each data entry annotated with 22 ecological characteristics such as feeding habits and activity patterns. The data is sourced from the iNaturalist and Xeno-canto platforms, and has undergone rigorous species screening and annotation validation assisted by GPT-5, and is released under the Creative Commons license. By integrating the biological taxonomic hierarchy, this dataset aims to enhance the recognition performance of unseen species, support interdisciplinary research on inferring ecological characteristics from acoustic signals, and provide a critical resource for biodiversity monitoring and computational ecology.

- 1AnimalCLAP: Taxonomy-Aware Language-Audio Pretraining for Species Recognition and Trait Inference大阪大学; 东京大学; 东京科学研究所 · 2026年



