▲ ▼ Deep Fake video detection
Deep Fake video is where the facial expression, speech of the person shown in the video aren't real and is computer generated using Machine Learning. This has profound consequences on how we validate the content, accelerating the fake news problem and even affecting the legal system in the countries which accept videos as evidence.
Imagine a video showing leader of a country discussing pre-emptive strike on another country, only that content of the video is not real but an ultra realistic deep fake video, but in the age of confirmation bias; we cannot expect people to validate everything they see or hear (if that was possible, fake news problem wouldn't exist in first place).
There are ethical use cases for deep fake video but there would be no stopping to exploitation of the deep fake tech for nefarious means. Hence there is a need gap for building tools which can be integrated in the systems for validating information.
A two year old paper titled 'Detecting Deep-Fake Videos from Phoneme-Viseme Mismatches' claims 80% detection of fakes by recognizing minute mismatches in the sounds people make and the shape of their mouths
New adversarial machine learning technique to leave traces of tampering when using editing tools - https://www.lightbluetouchpaper.org/2021/06/07/a-new-way-to-detect-deepfake-picture-editing/ .
Late, but commenting for posterity - Microsoft launches a deepfake detector tool ahead of US election - https://techcrunch.com/2020/09/02/microsoft-launches-a-deepfake-detector-tool-ahead-of-us-election .
Audio deepfakes are rising at an alarming fashion - https://www.nisos.com/white-papers/rise_synthetic_audio_deepfakes, so the need gap in deepfake detecting tech is not limited to just videos.
In fact, large scale frauds using audio deepfakes seems to be much more prevalent than video deepfakes due to lesser chances of victim recognising the fake content in time.
It's wise to inform our contacts that, if anyone with our voice sends a voice mail asking for money to be transferred it is not us.