DBD-research-group/WABAD

WABAD是一个用于被动声学监测的世界性标注鸟类声学数据集。原始数据集可在Zenodo上获得，包含了完整的作者列表。数据集的主要组成部分是大部分为一分钟的音频文件（.wav格式），以及对应的Audacity和Raven Pro标注文件。这个经过 curated 的版本对数据进行了调整以便于使用，包括将音频文件分割为3秒的片段、将多标签标注与地点元数据合并到Parquet文件中，以及预定义的训练和测试数据集分割（80/20比例）。分割方法是将Raven Pro标注的起始和结束时间转换为物种列表和一键编码，适用于每个3秒片段中至少有0.5秒的叫声。

WABAD is a worldwide annotated bird acoustic dataset for passive acoustic monitoring. The original dataset is available on Zenodo, including a complete list of authors. The main components of the dataset are mostly one-minute audio files (.wav format) and corresponding Audacity and Raven Pro annotation files. This curated version of the dataset has been adjusted for ease of use, including segmenting audio files into 3-second clips, combining multi-label annotations with site metadata in Parquet files, and predefined train and test dataset splits (80/20 ratio). The segmentation method converts Raven Pro annotation onset/offset times to a species list and one-hot encoding for 3-second segments where at least 0.5-second of vocalization is present.