FSD
收藏资源简介:
FSD数据集是由中国传媒大学媒体融合与传播国家重点实验室创建的,旨在研究歌曲深度伪造检测。该数据集包含200首真实歌曲和450首伪造歌曲,伪造歌曲通过五种先进的歌唱声音合成和转换技术生成。数据集的创建过程涉及从真实歌曲中提取乐器轨道,并与伪造的歌唱声音混合。FSD数据集主要应用于检测和验证歌曲的真实性,保护音乐作品的完整性,维护艺术家的声誉,并保持听众的信任。
The FSD dataset was created by the State Key Laboratory of Media Convergence and Communication, Communication University of China, aiming to research deepfake song detection. This dataset includes 200 authentic songs and 450 deepfake songs, which are generated via five cutting-edge singing voice synthesis and voice conversion technologies. The dataset development process involves extracting instrumental tracks from authentic songs and mixing them with the generated fake singing voices. The FSD dataset is primarily utilized for detecting and verifying the authenticity of songs, safeguarding the integrity of musical works, protecting artists' reputations, and maintaining listener trust.




