期刊名称：International Journal of Advances in Engineering and Management
电子版ISSN：2395-5252
出版年度：2020
卷号：2
期号：4
页码：860-865
DOI：10.35629/5252-0204821825
语种：English
出版社：IJAEM JOURNAL
摘要：The advent of deep learning generative models enables realistic generation from known data distribution, such as images, videos and sounds. Voice samples generated by such models can used for malicious purposes, i.e. fraud and impersonation if one fails to detect and report them. This poses challenges on the state-of-the-art voice verification systems to identify generated fake voices in order to prevent misuse of fake information. To test established verification systems against fake voices, we obtained a dataset of fake voices by CycleGAN-VC and used it to investigate two verification systems, 1) convolutional VAE, to see if they can detect generated fake voices.
