// Technology NEWS // With This AI, Make Anyone Say Anything

in #news5 years ago (edited)

Researchers have developed a technology that allows a filmed speech to be modified simply by editing the text of the verbal transcription. Fake news enthusiasts will love it.

Capture d’écran 2019-06-22 à 14.08.41.png

Source

Automatically editing a video interview has never been easier. Researchers at Princeton and Stanford Universities and the Max Planck Institute have just published a technique that allows the content of a video speech to be modified simply by editing the text of the verbal transcript.

The system is then able, from the new text, to create the corresponding video interview, without interruptions or jolts, as if it were original. This allows you to add and delete words, or simply rearrange them. The result is so amazing that most people who watch these modified videos think they are true.

Examples are shown in a presentation video on YouTube.

Other researchers are also exploring this field of automatic video generation. In 2017, a group from the University of Washington had succeeded in creating an artificial video of Barack Obama from an existing audio track.

In comparison, the technique just presented by the Princeton and Stanford researchers is superior, because it avoids the need to make an audio recording: all you have to do is edit a text.

How is this modification work carried out?

The system created by the researchers will first analyze the video and verbal transcription to recognize not only phonemes but also "vises". That is, the elementary facial expressions associated with phonemes. The system will then analyze the changes made to the text and identify the visions that should be used to modify the video.

A 3D modeling, then a neural network

The new sequences are then created in two steps. First, the software will generate a 3D model of the movements of the mouth and jaw, while respecting the general conditions of the video: exposure, lighting, head position, etc. In a second step, a previously trained neural network will transform these 3D models into realistic video sequences.

For the audio part, researchers use different methods: either they re-record the new text with the original person, or they artificially generate the new audio parts with software such as Adobe VoCo.


Source

Obviously, researchers do not want their technology to be used to maliciously manipulate speeches or interviews. They see it as an additional tool in audio-video production processes, for example, to correct recording errors. The software could also be used to generate realistic video sequences for a virtual assistant.

In all cases, the use of this technology should be open and transparent, and with the consent of the person being filmed. To avoid fraudulent and malicious creations, researchers believe that more verification techniques, such as forensic analysis or watermarking, should be developed.

Sources: Stanford's Project and YouTube

Stay Informed, Stay Safe

DQmdpsoEfLe5nRg4Q1oKWHNjLdMnAucCYfRou1yF5Yiwrzs.png

DQmNuF3L71zzxAyJB7Lk37yBqjBRo2uafTAudFDLzsoRV5L.gif

Sort:  

Congratulations @vijbzabyss! You have completed the following achievement on the Steem blockchain and have been rewarded with new badge(s) :

You published more than 250 posts. Your next target is to reach 300 posts.

You can view your badges on your Steem Board and compare to others on the Steem Ranking
If you no longer want to receive notifications, reply to this comment with the word STOP

To support your work, I also upvoted your post!

Do not miss the last post from @steemitboard:

The Steem community has lost an epic member! Farewell @woflhart!
SteemitBoard - Witness Update
Do not miss the coming Rocky Mountain Steem Meetup and get a new community badge!
Vote for @Steemitboard as a witness to get one more award and increased upvotes!

Congratulations! I just stopped by to say that your post has been selected as a daily Featured Post of my personal curation project! You can find the daily Featured Post HERE.
I upvoted your contribution and I put it on the list because to my mind your post is what I call a quality content!
I am @miti, a manual curator that shall make available all his Steem Power to authors deserving of support. Let's make STEEM great again!
Have a nice day and keep up the good work!

Coin Marketplace

STEEM 0.17
TRX 0.13
JST 0.028
BTC 57178.77
ETH 3095.71
USDT 1.00
SBD 2.12