Deepfakes aren’t only confined to videos. Deepfake audio is a rapidly expanding field with several uses. Realistic audio deepfakes can now be created using deep learning algorithms with as little as a few hours (or, in some cases, minutes) of the voice of the person whose audio is being cloned, and once a model of a voice is created, that person can be made to say anything, as was the case last year when the fake sound of a CEO was used to commit fraud.
Deepfake audio has medical uses in the original voice replacement, as well as in computer video games – now programmers may allow in-gamer characters to say anything in real-time, rather than depending on a restricted set of scripts produced before the game was released.
How to Identify Deepfake?
As deepfakes grow more popular, society will most likely need to adjust to identifying deepfake videos in the same way that internet consumers have become accustomed to detecting other types of fake news. In many cases, like in cybersecurity, more deep fake technology needs to arise to identify and prevent it from spreading, which may set off a vicious cycle and perhaps do even more harm.
Deepfakes can be identified by a few indicators:
· Current deepfakes have difficulty animating faces convincingly, resulting in videos in which the person never blinks, or blinks far too frequently or unnaturally. However, once University of Albany researchers published a paper revealing the blinking irregularity, fresh deepfakes were produced that no longer had this problem.
· Look for skin or hair issues, as well as faces that appear to be blurrier than the surroundings in which they are placed. The focus may appear abnormally soft.
· Is the illumination artificial? Deepfake algorithms frequently keep the lighting from the clips used as prototypes for the fake video, which would be a poor match for the brightness in the target video.
· The audio may not appear to fit the person, particularly if the video was fabricated but the actual audio was not as expertly edited.