Deep Fake: When AI simulates images and voices

AI produces amazingly real-looking images

It’s possible deep fake-Technology. Deepbecause “deep learning” is behind it, a special form of artificial intelligence. Learn here AI-Systems through extensive observation, for example how a person moves, what facial expressions they use, how they speak. After the “learning phase” such a system is able to imitate the person amazingly well – in a video. And “fake“—because it’s a fake.

Researchers have been working on this technology for years. The best known, because the first example that it works, is undoubtedly a Washington University project. In an official speech by Exde -US President Barack Obama, he said things that sound real and look real – but are a complete fake. Because even if Barack Obama secretly thinks so – he would probably never say publicly that “Donald Trump is a complete idiot”. in the deepfake– video but yes.

app: deep fakes as pop culture

smartphone-Users know quite a few today appwith which one “Face Swapping” can operate: mount your own face in a famous film sequence or your own child’s photo, which is suddenly animated – and into a current one song sings. That’s all deep fake-technologies. They have already arrived in pop culture today – they are so easy to use.

The image quality of deep fakes such app is mostly miserable. Because such images can be calculated faster – and you don’t see the possible errors. In order to raise the whole thing to the top level, more preparation time is needed, better AItechnology, more powerful servers and more effort overall. Not worth it for a 4 EUR app. For Hollywood or one deep fake-Attack on a political level, however.

This deepfake was laborious

And obviously we have exactly such a case here. The image quality of a video conference is not hollywood-like, but also generally bad. That makes it easier live such deep fakes to create.

Nevertheless, a lot of effort is necessary, because in a video conference, the fake-Images are generated live – including the spoken word. One can assume that this is not a (bad) schoolboy prank, but an elaborate one deep fake– Action behind which probably Russian forces are likely to be – only they are interested in the confusion associated with such an action.

Mimic voices: Alexa reads with grandma’s voice

Voices can also be included deepfake imitate. Manufacturer has been working for a long time Adobe on software that can mimic any voice. To do this, the AI just “listen” to 20 minutes of audio material from the person concerned. And she can software speak with a foreign voice – and synthesize speaking speed, pronunciation and intonation in a deceptively real way.

Also Amazon develops such technology. Just a few days ago, the group introduced an “Alexa” that can, for example, speak with the voice of the deceased grandmother – and read a bedtime story to the grandson. One minute of existing speech material should be enough to train the artificial voice. Another prototype, but quite possible that Amazon the technology will soon offer.

deep fakes are a danger

All these examples show: deep fakes can undoubtedly be very entertaining. In the cinema, in advertising, even on your own smartphone. However, they also harbor an enormous risk and threat potential. Because in the wrong hands and used on the political stage, they can have fatal consequences. Giffey’s team only became skeptical and suspicious after 15 minutes. Quite possible that this fake talk was just a first attempt – and the group behind it has even more plans.

#Deep #Fake #simulates #images #voices

Leave a Comment

Your email address will not be published.