Chinese Chorales Dataset: A High-Quality Music Dataset for Score Generation

For a long time, the JSB Chorales Dataset has served as the benchmark for choral composition generation, with numerous models and algorithms achieving remarkable results on this dataset, which is designed to generate Bach-style choral music. However, when we aim to tackle the task of generating Chinese vocal choral compositions, we encounter a lack of suitable Chinese music datasets for this purpose. The Chinese Chorales Dataset presented in this paper is a high-quality collection of Chinese choral music, comprising 125 Chinese choral songs stored in MusicXML format, divided into 441 musical segments. This dataset has been professionally crafted to meet the needs of Chinese composers seeking to create high-quality choral compositions. We also provide a compressed .npz file version containing pitch, fermata, tempo, and chord information, split into training, validation, and test sets. Additionally, we conducted multiple experiments on this dataset to validate the effectiveness of the information contained within. For access to the dataset and usage details, please visit https://github.com/123654ad/Chinese-Chorales-Dataset/tree/main .