CocoChorales

该数据集包含240,000个标准四声部合唱乐段，每个乐段为4/4拍，共8小节，使用了13种不同的乐器。每个样本是由这13种乐器中的四种乐器组合而成。数据集规模为：总计240,000个乐段，其中训练集192,000个，验证集24,000个，测试集24,000个。任务旨在训练和评估用于音乐生成和重建的模型。

This dataset contains 240,000 standard four-part chorale segments. Each segment follows a 4/4 time signature and spans 8 measures, with 13 distinct musical instruments available. Each sample is formed by combining four of these 13 instruments. The full dataset is split into three subsets: 192,000 segments for training, 24,000 for validation, and 24,000 for testing, totaling 240,000 segments. This dataset is intended for training and evaluating models for music generation and reconstruction.