Vocal-Accompaniment Pairs

该数据集包含了超过120万对时长为10秒的歌声和器乐音乐片段，这些片段经过筛选，以确保歌声和伴奏的清晰度。此外，数据集不仅包含了领域内的评估数据集，还包含了从MUSDB18数据集中衍生出的零样本评估数据集。总共有超过3000小时的音乐数据，其任务旨在生成歌唱伴奏。

This dataset contains over 1.2 million 10-second-long paired audio clips of singing vocals and instrumental music, which have been filtered to guarantee the clarity of both the singing vocals and their instrumental accompaniments. Furthermore, this dataset not only includes in-domain evaluation datasets, but also zero-shot evaluation datasets derived from the MUSDB18 dataset. In total, it encompasses over 3,000 hours of musical audio data, with the core task targeted at generating singing accompaniments.