dolphinteam/OpenWhistle-1.0-Detection-Finetuning

OpenWhistle-1.0-Detection-Finetuning是一个用于二进制哨声检测任务的数据集，标签分为noise和whistle两类。数据集总共有5600行数据，分为训练集（4480行）和测试集（1120行）。数据集的构建基于OpenWhistle检测剪辑的精选内容，通过将非noise的源标签合并为whistle来创建二进制标签。数据集的特征包括音频剪辑（以decode=False存储）、二进制类别标签、原始剪辑文件名和原始源标签。此外，数据集还提供了详细的标签计数、源标签计数和源组计数信息。

OpenWhistle-1.0-Detection-Finetuning is a dataset for binary whistle detection task, with labels noise and whistle. The dataset contains a total of 5600 rows, divided into a training set (4480 rows) and a test set (1120 rows). The dataset is constructed from curated OpenWhistle detection clips by collapsing every non-noise source label into whistle. Features of the dataset include audio clips (stored with decode=False), binary class labels, original clip filenames, and original source labels. Additionally, the dataset provides detailed counts of labels, source labels, and source groups.