Large-scale Unparalleled Singing Voice Dataset

该数据集是一个大规模的零样本歌声转换数据集，包含了超过1815小时的纯歌声以及6367位说话者。数据集中不仅包含了每首歌歌词及其时间信息，这些信息还经过了自动语音识别模型的处理并人工校对错误。该数据集规模宏大，拥有150万个音乐片段，涉及超过1万名说话者。其任务旨在实现零样本歌声转换。

This is a large-scale zero-shot singing voice conversion dataset, containing over 1815 hours of raw singing audio and 6367 unique speakers. The dataset includes the lyrics and timestamp information of each song, which have been processed by an automatic speech recognition model and manually proofread to correct errors. Boasting a massive scale, the dataset consists of 1.5 million music clips and covers more than 10,000 speakers in total. This dataset is specifically developed for zero-shot singing voice conversion tasks.