2015

Live DeepFake

Yisroel Mirksy, Oleg Brodt, Joshua Cohen, Ruven Levy, Itay Blokh

There are an increasing number of live deep fake platforms becoming readily available to the everyday user to easily swap or fake their own and others faces. These everyday users without background in technology or AI have at their fingertips the ability to easily swap and alter their faces. Snapchat for example has live face detection, swapping, and altering capabilities through their filters. FaceApp is another mobile platform that aids in quick face swapping with popular celebrities. Among its direct swapping capabilities, it can also alter the users faces, hair, skin tones and more, all in real time to change the appearance of the target user. Even minor altering such as changing skin tone and hair color, could have major implications when used irresponsibly or maliciously. These attacks are referred to as morphing attacks and identity fraud, and serve as a large threat to Border Protection and Photo ID verification. Next these can also be used in person to person interaction over the internet. One could alter their appearance and give a falso sense of who they are in real time. Following voice cloning and synthesis are becoming just as open and easy to use for the public. Open platforms such as Lyrebird and iSpeech, or even open GitHub repositories allow everyday users to clone voices on their own with little difficulty and repercussions. In this paper, well discuss not only the uses of this technology, but how we have used open source material to create a live synthesized fake of both audio and video in real time. Taking open source GitHub repositories on live deep fakes, and neural net voice cloning, we can create a fully functioning deep fake of both …