Deepfakes are manipulated image, audio or video documents in which products, people and actions have been intentionally inserted into the documents using artificial intelligence. In the documents, the contents are falsified in order to thereby circulate misrepresentations and discredit well-known personalities.
Deepfakes are falsifications of images and sounds. In the photos and video, faces and even entire persons are exchanged and imitated in the course of movement. While the exchange of faces in photos, face swapping, is still simple, videos in which the body posture, the movement dynamics, body puppetry, or the speech and accent adaptation, voice swapping, are much more complicated and computationally intensive. Hence, the use of artificial intelligence.
To perform the computationally intensive deepfake operations, two artificial neural networks, Generative Adversarial Networks( GAN), are used. The two networks work together, one forming a generator, the other a discriminator. The generator is a learning component whose results are generated according to a certain distribution. In contrast, the discriminator analyzes the results against the given distribution. Any deviations are adjusted by the discriminator.
Through the learning algorithm, the results presented are constantly improving.