DeepSpeak Dataset v1.0
收藏资源简介:
DeepSpeak Dataset v1.0是由加州大学伯克利分校和斯坦福大学联合创建的大规模音频视频数据集,旨在支持数字取证领域的研究。该数据集包含6226条记录,涵盖220名参与者的真实和深度伪造视频,总计34小时。数据集通过Prolific研究招募平台收集,确保了性别和种族的多样性。创建过程中,参与者通过自定义的Google Chrome应用录制视频,随后进行音频和视频的合成与处理。该数据集主要用于开发和评估深度伪造检测技术,支持学术研究和非商业用途。
The DeepSpeak Dataset v1.0 is a large-scale audio-visual dataset jointly developed by the University of California, Berkeley and Stanford University, intended to support research in the field of digital forensics. This dataset contains 6,226 records, covering both authentic and deepfake videos from 220 participants, with a total duration of 34 hours. It was collected via the Prolific research recruitment platform, which ensures diversity in gender and ethnicity. During the dataset creation process, participants recorded videos using a custom Google Chrome application, followed by audio and video synthesis and processing. This dataset is primarily used for developing and evaluating deepfake detection technologies, and supports academic research and non-commercial use.

- 1DeepSpeak Dataset v1.0加州大学伯克利分校 · 2024年



