MusicBench

该数据集包含了5,000个五秒钟的片段，这些片段源自3,413首高质量的音乐作品，涵盖了不同的音乐风格、类型和乐器配置。此外，该数据集还作为评估的基准真值和基线。规模达到了5,000个片段，任务是对个性化音乐生成中的噪音混合进行研究。

This dataset consists of 5,000 5-second audio clips extracted from 3,413 high-quality musical compositions, spanning a wide range of music genres, styles and instrument configurations. Additionally, this dataset serves as the ground truth and baseline for model evaluation. With a total of 5,000 clips, this dataset is tailored for research on noise mixing in personalized music generation.