SingFake
收藏资源简介:
该数据集名为SingFake,包含了40位真实歌手及其对应的AI歌手,涵盖了多种语言,共有634首真实歌曲和671首深度伪造歌曲,这些歌曲真实地反映了在线社区中普遍存在的深度伪造细微差别。该数据集被划分为训练集、验证集和测试集,并设有特定的测试子集(T01-T04),旨在评估歌唱声音深度伪造检测系统的鲁棒性。规模上,数据集包含了634首真实歌曲和671首深度伪造歌曲,其任务专注于歌唱声音深度伪造的检测。
This dataset is named SingFake. It encompasses 40 real singers and their corresponding AI-synthesized singing voices, spanning multiple languages. The dataset comprises 634 authentic songs and 671 deepfake songs, which authentically capture the prevalent nuances of deepfake singing voices in online communities. It is split into training, validation, and test sets, with dedicated test subsets (T01-T04) designed to evaluate the robustness of singing voice deepfake detection systems. The core task of this dataset focuses on singing voice deepfake detection.




