Synthetic human-like fakes

From Stop Synthetic Filth! wiki
Revision as of 11:48, 17 July 2023 by Juho Kunsola (talk | contribs) (→‎2020's synthetic human-like fakes: + [...] made with a stolen messaging service account)
Jump to navigation Jump to search

Definitions

When the camera does not exist, but the subject being imaged with a simulation of a (movie) camera deceives the watcher to believe it is some living or dead person it is a digital look-alike.

In 2017-2018 this started to be referred to as w:deepfake, even though altering video footage of humans with a computer with a deceiving effect is actually 20 yrs older than the name "deep fakes" or "deepfakes".[1][2]

When it cannot be determined by human testing or media forensics whether some fake voice is a synthetic fake of some person's voice, or is it an actual recording made of that person's actual real voice, it is a pre-recorded digital sound-alike. This is now commonly referred to as w:audio deepfake.

Real-time digital look-and-sound-alike in a video call was used to defraud a substantial amount of money in 2023.[3]

Read more about synthetic human-like fakes, see and support organizations and events against synthetic human-like fakes and what they are doing, what kinds of Laws against synthesis and other related crimes have been formulated, examine the SSFWIKI timeline of synthetic human-like fakes or view the Mediatheque.


This is not a picture of Obama, because it is not Obama in the video that this screenshot is from, but a synthetic human-like fake, more precisely a pre-recorded digital look-alike.

Click on the picture or Obama's appearance thieved - a public service announcement digital look-alike by Monkeypaw Productions and Buzzfeed to view an April 2018 public service announcement moving digital look-alike made to appear Obama-like. The video is accompanied with imitator sound-alike, and was made by w:Monkeypaw Productions (.com) in conjunction with w:BuzzFeed (.com). You can also View the same video at YouTube.com.[4]
Image 2 (low resolution rip) shows a 1999 technique for sculpting a morphable model, till it matches the target's appearance.
(1) Sculpting a morphable model to one single picture
(2) Produces 3D approximation
(4) Texture capture
(3) The 3D model is rendered back to the image with weight gain
(5) With weight loss
(6) Looking annoyed
(7) Forced to smile Image 2 by Blanz and Vettel – Copyright ACM 1999 – http://dl.acm.org/citation.cfm?doid=311535.311556 – Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page.

Digital look-alikes

It is recommended that you watch In Event of Moon Disaster - FULL FILM (2020) at the moondisaster.org project website (where it has interactive portions) by the Center for Advanced Virtuality of the w:MIT


Introduction to digital look-alikes

Image 1: Separating specular and diffuse reflected light

(a) Normal image in dot lighting

(b) Image of the diffuse reflection which is caught by placing a vertical polarizer in front of the light source and a horizontal in the front the camera

(c) Image of the highlight specular reflection which is caught by placing both polarizers vertically

(d) Subtraction of c from b, which yields the specular component

Images are scaled to seem to be the same luminosity.

Original image by Debevec et al. – Copyright ACM 2000 – https://dl.acm.org/citation.cfm?doid=311779.344855 – Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page.
Subtraction of the diffuse reflection from the specular reflection yields the specular component of the model's reflectance.

Original picture by w:Paul Debevec et al. - Copyright ACM 2000 https://dl.acm.org/citation.cfm?doid=311779.344855

In the cinemas we have seen digital look-alikes for over 15 years. These digital look-alikes have "clothing" (a simulation of clothing is not clothing) or "superhero costumes" and "superbaddie costumes", and they don't need to care about the laws of physics, let alone laws of physiology. It is generally accepted that digital look-alikes made their public debut in the sequels of The Matrix i.e. w:The Matrix Reloaded and w:The Matrix Revolutions released in 2003. It can be considered almost certain, that it was not possible to make these before the year 1999, as the final piece of the puzzle to make a (still) digital look-alike that passes human testing, the reflectance capture over the human face, was made for the first time in 1999 at the w:University of Southern California and was presented to the crème de la crème of the computer graphics field in their annual gathering SIGGRAPH 2000.[5]


“Do you think that was w:Hugo Weaving's left cheekbone that w:Keanu Reeves punched in with his right fist?”

~ Trad on The Matrix Revolutions



The problems with digital look-alikes

Extremely unfortunately for the humankind, organized criminal leagues, that posses the weapons capability of making believable looking synthetic pornography, are producing on industrial production pipelines terroristic synthetic pornography[footnote 1] by animating digital look-alikes and distributing it in the murky Internet in exchange for money stacks that are getting thinner and thinner as time goes by.

These industrially produced pornographic delusions are causing great human suffering, especially in their direct victims, but they are also tearing our communities and societies apart, sowing blind rage, perceptions of deepening chaos, feelings of powerlessness and provoke violence. These kinds of hate illustration increases and strengthens hate feeling, hate thinking, hate speech and hate crimes and tears our fragile social constructions apart and with time perverts humankind's view of humankind into an almost unrecognizable shape, unless we interfere with resolve.

Fixing the problems from digital look-alikes

We need to move on 3 fields: legal, technological and cultural.

Technological: Computer vision system like FacePinPoint.com for seeking unauthorized pornography / nudes used to exist 2017-2021 and could be revived if funding is found. It was a service practically identical with SSFWIKI original concept Adequate Porn Watcher AI (concept).

Legal: Legislators around the planet have been waking up to this reality that not everything that seems a video of people is a video of people and various laws have been passed to protect humans and humanity from the menaces of synthetic human-like fakes, mostly digital look-alikes so far, but hopefully humans will be protected also fro other aspects of synthetic human-like fakes by laws. See Laws against synthesis and other related crimes

Age analysis and rejuvenating and aging syntheses

Temporal limit of digital look-alikes

A picture of the 1895 w:Cinematograph

w:History of film technology has information about where the border is.

Digital look-alikes cannot be used to attack people who existed before the technological invention of film. For moving pictures the breakthrough is attributed to w:Auguste and Louis Lumière's w:Cinematograph premiered in Paris on 28 December 1895, though this was only the commercial and popular breakthrough, as even earlier moving pictures exist. (adapted from w:History of film)

The w:Kinetoscope is an even earlier motion picture exhibition device. A prototype for the Kinetoscope was shown to a convention of the National Federation of Women's Clubs on May 20, 1891.[6] The first public demonstration of the Kinetoscope was held at the Brooklyn Institute of Arts and Sciences on May 9, 1893. (Wikipedia)[6]


Digital sound-alikes

University of Florida published an antidote to synthetic human-like fake voices in 2022

2022 saw a brilliant counter-measure presented to peers at the 31st w:USENIX Security Symposium 10-12 August 2022 by w:University of Florida Who Are You (I Really Wanna Know)? Detecting Audio DeepFakes Through Vocal Tract Reconstruction.

The university's foundation has applied for a patent and let us hope that they will w:copyleft the patent as this protective method needs to be rolled out to protect the humanity.

Below transcluded from the article

Detecting deep-fake audio through vocal tract reconstruction is an epic scientific work, against fake human-like voices, from the w:University of Florida in published to peers in August 2022.

University of Florida, the home of Florida Institute for Cybersecurity Research (FICS) at fics.institute.ufl.edu, that brought us the Detecting deep-fake audio through vocal tract reconstruction system to hunt down the fake human-like voices.
The Office of Naval Research (ONR) at nre.navy.mil of the USA funded this breakthrough science.

The work Who Are You (I Really Wanna Know)? Detecting Audio DeepFakes Through Vocal Tract Reconstruction at usenix.org, presentation page, version included in the proceedings[7] and slides from researchers of the Florida Institute for Cybersecurity Research (FICS) at fics.institute.ufl.edu in the w:University of Florida received funding from the w:Office of Naval Research and was presented on 2022-08-11 at the 31st w:USENIX Security Symposium.

This work was done by PhD student Logan Blue, Kevin Warren, Hadi Abdullah, Cassidy Gibson, Luis Vargas, Jessica O’Dell, Kevin Butler and Professor Patrick Traynor.

The University of Florida Research Foundation Inc has filed for and received an US patent titled 'Detecting deep-fake audio through vocal tract reconstruction' registration number US20220036904A1 (link to patents.google.com) with 20 claims. The patent application was published on Thursday 2022-02-03. The patent application was approved on 2023-07-04 and has an adjusted expiration date of Sunday 2041-12-29.PhD student Logan Blue and professor Patrick Traynor wrote an article for the general public on the work titled Deepfake audio has a tell – researchers use fluid dynamics to spot artificial imposter voices at theconversation.com[8] that was published Tuesday 2022-09-20 and permanently w:copylefted it under Creative Commons Attribution-NoDerivatives (CC-BY-ND).

This new counter-measure needs to be rolled out to humans to protect humans against the fake human-like voices.


Below is an exact copy of the original article from the SSF! wordpress titled "Amazing method and results from University of Florida scientists in 2022 against the menaces of digital sound-alikes / audio deepfakes". Thank you to the original writers for having the wisdom of licensing the article under CC-BY-ND.


On known history of digital sound-alikes

A picture of a cut-away titled "Voice-terrorist could mimic a leader" from a 2012 w:Helsingin Sanomat warning that the sound-like-anyone machines are approaching. Thank you to homie Prof. David Martin Howard of the w:University of York, UK and the anonymous editor for the heads-up.

The first English speaking digital sound-alikes were first introduced in 2016 by Adobe and Deepmind, but neither of them were made publicly available.

Then in 2018 at the w:Conference on Neural Information Processing Systems (NeurIPS) the work 'Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis' (at arXiv.org) was presented. The pre-trained model is able to steal voices from a sample of only 5 seconds with almost convincing results

The Iframe below is transcluded from 'Audio samples from "Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis"' at google.gituhub.io, the audio samples of a sound-like-anyone machine presented as at the 2018 w:NeurIPS conference by Google researchers.

Have a listen.

Observe how good the "VCTK p240" system is at deceiving to think that it is a person that is doing the talking.


Reporting on the sound-like-anyone-machines


The to the right video 'This AI Clones Your Voice After Listening for 5 Seconds' by '2 minute papers' at YouTube describes the voice thieving machine presented by Google Research in w:NeurIPS 2018.

Video video 'This AI Clones Your Voice After Listening for 5 Seconds' by '2 minute papers' at YouTube describes the voice thieving machine by Google Research in w:NeurIPS 2018.

Documented crimes with digital sound-alikes

In 2019 reports of crimes being committed with digital sound-alikes started surfacing. As of Jan 2022 no reports of other types of attack than fraud have been found.

2019 digital sound-alike enabled fraud

By 2019 digital sound-alike anyone technology found its way to the hands of criminals. In 2019 Symantec researchers knew of 3 cases where digital sound-alike technology had been used for w:crime.[9]

Of these crimes the most publicized was a fraud case in March 2019 where 220,000€ were defrauded with the use of a real-time digital sound-alike.[10] The company that was the victim of this fraud had bought some kind of cyberscam insurance from French insurer w:Euler Hermes and the case came to light when Mr. Rüdiger Kirsch of Euler Hermes informed w:The Wall Street Journal about it.[11]

Reporting on the 2019 digital sound-alike enabled fraud

2020 digital sound-alike fraud attempt

In June 2020 fraud was attempted with a poor quality pre-recorded digital sound-alike with delivery method was voicemail. (Listen to a redacted clip at soundcloud.com) The recipient in a tech company didn't believe the voicemail to be real and alerted the company and they realized that someone tried to scam them. The company called in Nisos to investigate the issue. Nisos analyzed the evidence and they were certain it was a fake, but had aspects of a cut-and-paste job to it. Nisos prepared a report titled "The Rise of Synthetic Audio Deepfakes" at nisos.com on the issue and shared it with Motherboard, part of w:Vice (magazine) prior to its release.[13]

2021 digital sound-alike enabled fraud

The 2nd publicly known fraud done with a digital sound-alike[1st seen in 1] took place on Friday 2021-01-15. A bank in Hong Kong was manipulated to wire money to numerous bank accounts by using a voice stolen from one of the their client company's directors. They managed to defraud $35 million of the U.A.E. based company's money.[14]. This case came into light when Forbes saw a document where the U.A.E. financial authorities were seeking administrative assistance from the US authorities towards the end of recovering a small portion of the defrauded money that had been sent to bank accounts in the USA.[14]

Reporting on the 2021 digital sound-alike enabled fraud


Example of a hypothetical 4-victim digital sound-alike attack

A very simple example of a digital sound-alike attack is as follows:

Someone puts a digital sound-alike to call somebody's voicemail from an unknown number and to speak for example illegal threats. In this example there are at least two victims:

  1. Victim #1 - The person whose voice has been stolen into a covert model and a digital sound-alike made from it to frame them for crimes
  2. Victim #2 - The person to whom the illegal threat is presented in a recorded form by a digital sound-alike that deceptively sounds like victim #1
  3. Victim #3 - It could also be viewed that victim #3 is our law enforcement systems as they are put to chase after and interrogate the innocent victim #1
  4. Victim #4 - Our judiciary which prosecutes and possibly convicts the innocent victim #1.

Examples of speech synthesis software not quite able to fool a human yet

Some other contenders to create digital sound-alikes are though, as of 2019, their speech synthesis in most use scenarios does not yet fool a human because the results contain tell tale signs that give it away as a speech synthesizer.

Temporal limit of digital sound-alikes

w:Thomas Edison and his early w:phonograph. Cropped from w:Library of Congress copy, ca. 1877, (probably 18 April 1878)

The temporal limit of whom, dead or living, the digital sound-alikes can attack is defined by the w:history of sound recording.

The article starts by mentioning that the invention of the w:phonograph by w:Thomas Edison in 1877 is considered the start of sound recording.

The phonautograph is the earliest known device for recording w:sound. Previously, tracings had been obtained of the sound-producing vibratory motions of w:tuning forks and other objects by physical contact with them, but not of actual sound waves as they propagated through air or other media. Invented by Frenchman W:Édouard-Léon Scott de Martinville, it was patented on March 25, 1857.[15]

Apparently, it did not occur to anyone before the 1870s that the recordings, called phonautograms, contained enough information about the sound that they could, in theory, be used to recreate it. Because the phonautogram tracing was an insubstantial two-dimensional line, direct physical playback was impossible in any case. Several phonautograms recorded before 1861 were successfully played as sound in 2008 by optically scanning them and using a computer to process the scans into digital audio files. (Wikipedia)

A w:spectrogram of a male voice saying 'nineteenth century'

What should we do about digital sound-alikes?

Living people can defend[footnote 2] themselves against digital sound-alike by denying the things the digital sound-alike says if they are presented to the target, but dead people cannot. Digital sound-alikes offer criminals new disinformation attack vectors and wreak havoc on provability.

For these reasons the bannable raw materials i.e. covert voice models should be prohibited by law in order to protect humans from abuse by criminal parties.

It is high time to act and to criminalize the covert modeling of human voice!

Text syntheses

w:Chatbots and w:spamming have existed for a longer time, but only now armed with AI they are becoming more deceiving.

In w:natural language processing development in w:natural-language understanding leads to more cunning w:natural-language generation AI.

w:Large language models (LLM) are very large w:language models consisting of a w:neural network with many parameters.

w:OpenAI's w:Generative Pre-trained Transformer (GPT) is a left-to-right w:transformer (machine learning model)-based text generation model succeeded by w:GPT-2 and w:GPT-3

November 2022 saw the publication of OpenAI's w:ChatGPT, a conversational artificial intelligence.

w:Bard (chatbot) is a conversational w:generative artificial intelligence w:chatbot developed by w:Google, based on the w:LaMDA family of w:large language models. It was developed as a direct response to the rise of w:OpenAI's w:ChatGPT, and was released in March 2023. (Wikipedia)

Reporting / announcements (in reverse chronology)

External links

Detectors for synthesized texts

Introduction of w:ChatGPT by OpenAI brought the need for software to detect machine-generated texts.

Try AI plagiarism detection for free

For-a-fee AI plagiarism detection tools

Handwriting syntheses

Handwriting syntheses could be used

  1. Defensively, to hide one's handwriting style from public view
  2. Offensively, to thieve somebody else's handwriting style

If the handwriting-like synthesis passes human and media forensics testing, it is a digital handwrite-alike.

Here we find a risk similar to that which realized when the w:speaker recognition systems turned out to be instrumental in the development of digital sound-alikes. After the knowledge needed to recognize a speaker was w:transferred into a generative task in 2018 by Google researchers, we no longer cannot effectively determine for English speakers which recording is human in origin and which is from a machine origin.

Handwriting-like syntheses: w:Recurrent neural networks (RNN) seem are a popular choice for this task.


  1. Recurrent neural network handwriting generation demo at cs.toronto.edu is a demonstration site for publication
  2. Calligrapher.ai - Realistic computer-generated handwriting - The user may control parameters: speed, legibility, stroke width and style. The domain is registered by some organization in Iceland and the website offers no about-page[1st seen in 4]. According to this reddit post Calligrapher.ai is based on Graves' 2013 work, but "adds an w:inference model to allow for sampling latent style vectors (similar to the VAE model used by SketchRNN)".[17]

Handwriting recognition

Singing syntheses

As of 2020 the digital sing-alikes may not yet be here, but when we hear a faked singing voice and we cannot hear that it is fake, then we will know. An ability to sing does not seem to add much hostile capabilities compared to the ability to thieve spoken word.


Timeline of synthetic human-like fakes

See the #SSFWIKI Mediatheque for viewing media that is or is probably to do with synthetic human-like fakes.

2020's synthetic human-like fakes

  • 2023 | Real-time digital look-and-sound-alike crime | In April a man in northern China was defrauded of 4.3 million yuan by a criminal employing a digital look-and-sound-alike pretending to be his friend on a video call made with a stolen messaging service account.[3]
    • "Ahead of the election in Turkey, President Recep Tayyip Erdogan showed a video linking his main challenger Kemal Kilicdaroglu to the militant Kurdish organization PKK." [...] "Research by DW's fact-checking team in cooperation with DW's Turkish service shows that the video at the campaign rally was manipulated by combining two separate videos with totally different backgrounds and content." reports dw.com
  • 2023 | January 1st | Law | Law on sexual offences in Finland 2023 is found in Chapter 20 of the Finnish Criminal Code titled "Seksuaalirikoksista" ("Sexual offences") and came into effect on Sunday 2023-01-01.[20]

The new law in Finland protects adults against sexual image based abuse be it real or synthetic in origin.

Other countries have also woken up to the problems of synthesis crime and have legislated laws against synthesis and other related crimes.

Relevant sections of Chapter 20


This 2023 upgrade and gather-together of the Finnish Criminal Code on sexual offences was made upon the initiative of the 2019-2023 w:Marin Cabinet, was voted into law by the w:Members of the Parliament of Finland, 2019–2023 and it came into effect on Sunday 2023-01-01.

Translation to English by the Ministry of Justice: Criminal Code (39/1889) - Chapter 20 - Sexual offences (translation) as .pdf at oikeusministerio.fi (subject to possible revisions)

University of Florida, the home of Florida Institute for Cybersecurity Research (FICS) at fics.institute.ufl.edu, that brought us the Detecting deep-fake audio through vocal tract reconstruction system to hunt down the fake human-like voices.
The Office of Naval Research (ONR) at nre.navy.mil of the USA funded this breakthrough science.

The work Who Are You (I Really Wanna Know)? Detecting Audio DeepFakes Through Vocal Tract Reconstruction at usenix.org, presentation page, version included in the proceedings[7] and slides from researchers of the Florida Institute for Cybersecurity Research (FICS) at fics.institute.ufl.edu in the w:University of Florida received funding from the w:Office of Naval Research and was presented on 2022-08-11 at the 31st w:USENIX Security Symposium.

This work was done by PhD student Logan Blue, Kevin Warren, Hadi Abdullah, Cassidy Gibson, Luis Vargas, Jessica O’Dell, Kevin Butler and Professor Patrick Traynor.

The University of Florida Research Foundation Inc has filed for and received an US patent titled 'Detecting deep-fake audio through vocal tract reconstruction' registration number US20220036904A1 (link to patents.google.com) with 20 claims. The patent application was published on Thursday 2022-02-03. The patent application was approved on 2023-07-04 and has an adjusted expiration date of Sunday 2041-12-29.

PhD student Logan Blue and professor Patrick Traynor wrote an article for the general public on the work titled Deepfake audio has a tell – researchers use fluid dynamics to spot artificial imposter voices at theconversation.com[8] that was published Tuesday 2022-09-20 and permanently w:copylefted it under Creative Commons Attribution-NoDerivatives (CC-BY-ND).. Presented to peers in August 2022 and to the general public in September 2022.
  • 2022 | science | w:DALL-E 2, a successor designed to generate more realistic images at higher resolutions that "can combine concepts, attributes, and styles" was published in April 2022.[21] (Wikipedia)


  • 2021 | crime / fraud | The 2nd publicly known fraud done with a digital sound-alike[1st seen in 1] took place on Friday 2021-01-15. A bank in Hong Kong was manipulated to wire money to numerous bank accounts by using a voice stolen from one of the their client company's directors. They managed to defraud $35 million of the U.A.E. based company's money.[14]. This case came into light when Forbes saw a document where the U.A.E. financial authorities were seeking administrative assistance from the US authorities towards the end of recovering a small portion of the defrauded money that had been sent to bank accounts in the USA.[14]

Reporting on the 2021 digital sound-alike enabled fraud


  • 2021 | science and demonstration | DALL-E, a w:deep learning model developed by w:OpenAI to generate digital images from w:natural language descriptions, called "prompts" was published in January 2021. DALL-E uses a version of w:GPT-3 modified to generate images. (Adapted from Wikipedia)
In Dec 2020 Channel 4 aired a Queen-like fake i.e. they had thieved the appearance of Queen Elizabeth II using deepfake methods.
  • 2020 | Controversy / Public service announcement | Channel 4 thieved the appearance of Queen Elizabeth II using deepfake methods. The product of synthetic human-like fakery originally aired on Channel 4 on 25 December at 15:25 GMT.[26] View in YouTube
  • 2020 | Chinese legislation | On Wednesday January 1 2020 Chinese law requiring that synthetically faked footage should bear a clear notice about its fakeness came into effect. Failure to comply could be considered a w:crime the w:Cyberspace Administration of China (cac.gov.cn) stated on its website. China announced this new law in November 2019.[30] The Chinese government seems to be reserving the right to prosecute both users and w:online video platforms failing to abide by the rules. [31]


2010's synthetic human-like fakes


Code of Virginia (TOC) » Title 18.2. Crimes and Offenses Generally » Chapter 8. Crimes Involving Morals and Decency » Article 5. Obscenity and Related Offenses » Section § 18.2-386.2. Unlawful dissemination or sale of images of another; penalty

The section § 18.2-386.2. Unlawful dissemination or sale of images of another; penalty. of Virginia is as follows:

A. Any w:person who, with the w:intent to w:coerce, w:harass, or w:intimidate, w:maliciously w:disseminates or w:sells any videographic or still image created by any means whatsoever that w:depicts another person who is totally w:nude, or in a state of undress so as to expose the w:genitals, pubic area, w:buttocks, or female w:breast, where such person knows or has reason to know that he is not w:licensed or w:authorized to disseminate or sell such w:videographic or w:still image is w:guilty of a Class 1 w:misdemeanor.

For purposes of this subsection, "another person" includes a person whose image was used in creating, adapting, or modifying a videographic or still image with the intent to depict an actual person and who is recognizable as an actual person by the person's w:face, w:likeness, or other distinguishing characteristic.

B. If a person uses w:services of an w:Internet service provider, an electronic mail service provider, or any other information service, system, or access software provider that provides or enables computer access by multiple users to a computer server in committing acts prohibited under this section, such provider shall not be held responsible for violating this section for content provided by another person.

C. Venue for a prosecution under this section may lie in the w:jurisdiction where the unlawful act occurs or where any videographic or still image created by any means whatsoever is produced, reproduced, found, stored, received, or possessed in violation of this section.

D. The provisions of this section shall not preclude prosecution under any other w:statute.[35]

The identical bills were House Bill 2678 presented by w:Delegate w:Marcus Simon to the w:Virginia House of Delegates on January 14 2019 and three day later an identical Senate bill 1736 was introduced to the w:Senate of Virginia by Senator w:Adam Ebbin.

  • 2019 | demonstration | 'Thispersondoesnotexist.com' (since February 2019) by Philip Wang. It showcases a w:StyleGAN at the task of making an endless stream of pictures that look like no-one in particular, but are eerily human-like. Relevancy: certain
w:Google's logo. Google Research demonstrated their sound-like-anyone-machine at the 2018 w:Conference on Neural Information Processing Systems (NeurIPS). It requires only 5 seconds of sample to steal a voice.
  • 2018 | controversy / demonstration | The w:deepfakes controversy surfaces where porn videos were doctored utilizing w:deep machine learning so that the face of the actress was replaced by the software's opinion of what another persons face would look like in the same pose and lighting.
w:Adobe Inc.'s logo. We can thank Adobe for publicly demonstrating their sound-like-anyone-machine in 2016 before an implementation was sold to criminal organizations.
#w:Adobe Voco. Adobe Audio Manipulator Sneak Peak with w:Jordan Peele (at Youtube.com). November 2016 demonstration of a Adobe's unreleased sound-like-anyone-machine, the w:Adobe Voco at the w:Adobe MAX 2016 event in w:San Diego, w:California. The original Adobe Voco required 20 minutes of sample to thieve a voice.
  • 2013 | demonstration | At the 2013 SIGGGRAPH w:Activision and USC presented a w:real time computing "Digital Ira" a digital face look-alike of Ari Shapiro, an ICT USC research scientist,[42] utilizing the USC light stage X by Ghosh et al. for both reflectance field and motion capture.[43] The end result both precomputed and real-time rendering with the modernest game w:GPU shown here and looks fairly realistic.

2000's synthetic human-like fakes

  • 2009 | movie | A digital look-alike of a younger w:Arnold Schwarzenegger was made for the movie w:Terminator Salvation though the end result was critiqued as unconvincing. Facial geometry was acquired from a 1984 mold of Schwarzenegger.
  • 2009 | demonstration | Paul Debevec: 'Animating a photo-realistic face' at ted.com Debevec et al. presented new digital likenesses, made by w:Image Metrics, this time of actress w:Emily O'Brien whose reflectance was captured with the USC light stage 5. At 00:04:59 you can see two clips, one with the real Emily shot with a real camera and one with a digital look-alike of Emily, shot with a simulation of a camera - Which is which is difficult to tell. Bruce Lawmen was scanned using USC light stage 6 in still position and also recorded running there on a w:treadmill. Many, many digital look-alikes of Bruce are seen running fluently and natural looking at the ending sequence of the TED talk video. [44] Motion looks fairly convincing contrasted to the clunky run in the w:Animatrix: Final Flight of the Osiris which was w:state-of-the-art in 2003 if photorealism was the intention of the w:animators.
Logo of the w:The Matrix (franchise)
Traditional w:BRDF vs. subsurface scattering inclusive BSSRDF i.e. w:Bidirectional scattering-surface reflectance distribution function.

An analytical BRDF must take into account the subsurface scattering, or the end result will not pass human testing.
Music video for Bullet by w:Covenant from 2002. Here you can observe the classic "skin looks like cardboard"-bug that stopped the pre-reflectance capture era versions from passing human testing.
  • 2002 | music video | 'Bullet' by Covenant on Youtube by w:Covenant (band) from their album w:Northern Light (Covenant album). Relevancy: Contains the best upper-torso digital look-alike of Eskil Simonsson (vocalist) that their organization could procure at the time. Here you can observe the classic "skin looks like cardboard"-bug (assuming this was not intended) that thwarted efforts to make digital look-alikes that pass human testing before the reflectance capture and dissection in 1999 by w:Paul Debevec et al. at the w:University of Southern California and subsequent development of the "Analytical w:BRDF" (quote-unquote) by ESC Entertainment, a company set up for the sole purpose of making the cinematography for the 2003 films Matrix Reloaded and Matrix Revolutions possible, lead by George Borshukov.

1990's synthetic human-like fakes

Logo of the w:Institute for Creative Technologies founded in 1999 in the w:University of Southern California by the w:United States Army

1970's synthetic human-like fakes

w:A Computer Animated Hand is a 1972 short film by w:Edwin Catmull and w:Fred Parke. This was the first time that w:computer-generated imagery was used in film to animate likenesses of moving human appearance.
  • 1976 | movie | w:Futureworld reused parts of A Computer Animated Hand on the big screen.

1960's synthetic human-like fakes

1930's synthetic human-like fakes

w:Voder demonstration pavillion at the w:1939 New York World's Fair

1770's synthetic human-like fakes

A replica of w:Wolfgang von Kempelen's w:Wolfgang von Kempelen's Speaking Machine, built 2007–09 at the Department of w:Phonetics, w:Saarland University, w:Saarbrücken, Germany. This machine added models of the tongue and lips, enabling it to produce w:consonants as well as w:vowels

Footnotes

  1. ↑ It is terminologically more precise, more inclusive and more useful to talk about 'terroristic synthetic pornography', if we want to talk about things with their real names, than 'synthetic rape porn', because also synthesizing recordings of consentual looking sex scenes can be terroristic in intent.
  2. ↑ Whether a suspect can defend against faked synthetic speech that sounds like him/her depends on how up-to-date the judiciary is. If no information and instructions about digital sound-alikes have been given to the judiciary, they likely will not believe the defense of denying that the recording is of the suspect's voice.

Contact information of organizations

Please contact these organizations and tell them to work harder against the disinformation weapons


1st seen in

  1. ↑ 1.0 1.1 1.2 1.3 https://www.reddit.com/r/VocalSynthesis/
  2. ↑ 2.0 2.1 2.2 2.3 2.4 2.5 https://wordlift.io/blog/en/best-plagiarism-checkers-for-ai-generated-content/
  3. ↑ https://www.ucl.ac.uk/news/2016/aug/new-computer-programme-replicates-handwriting via Google search for "ai handwriting generator"
  4. ↑ https://seanvasquez.com/handwriting-generation redirects to Calligrapher.ai - seen in https://www.reddit.com/r/MachineLearning/comments/gh9cbg/p_generate_handwriting_with_an_inbrowser/
  5. ↑ Witness newsletter I subscribed to at https://www.witness.org/get-involved/
  6. ↑ 6.0 6.1 https://www.connectedpapers.com/main/8fc09dfcff78ac9057ff0834a83d23eb38ca198a/Transfer-Learning-from-Speaker-Verification-to-Multispeaker-TextToSpeech-Synthesis/graph
  7. ↑ https://www.technologyreview.com/2020/08/28/1007746/ai-deepfakes-memes/
  8. ↑ 'US Lawmakers: AI-Generated Fake Videos May Be a Security Threat' at uk.pcmag.com, 2018-09-13 reporting by Michael Kan
  9. ↑ PROTECTING PRESIDENT ZELENSKYY AGAINST DEEP FAKES https://arxiv.org/pdf/2206.12043.pdf


References

  1. ↑ 1.0 1.1 Boháček, Matyáš; Farid, Hany (2022-11-23). "Protecting world leaders against deep fakes using facial, gestural, and vocal mannerisms". w:Proceedings of the National Academy of Sciences of the United States of America. 119 (48). doi:10.1073/pnas.221603511. Retrieved 2023-01-05.
  2. ↑ 2.0 2.1 Bregler, Christoph; Covell, Michele; Slaney, Malcolm (1997-08-03). "Video Rewrite: Driving Visual Speech with Audio" (PDF). SIGGRAPH '97: Proceedings of the 24th annual conference on Computer graphics and interactive techniques: 353–360. doi:10.1145/258734.258880. Retrieved 2022-09-09.
  3. ↑ 3.0 3.1 "'Deepfake' scam in China fans worries over AI-driven fraud". w:Reuters.com. w:Reuters. 2023-05-22. Retrieved 2023-06-05.
  4. ↑ "You Won't Believe What Obama Says In This Video!". w:YouTube. w:BuzzFeed. 2018-04-17. Retrieved 2022-01-05. We're entering an era in which our enemies can make anyone say anything at any point in time.
  5. ↑ 5.0 5.1 Debevec, Paul (2000). "Acquiring the reflectance field of a human face". Proceedings of the 27th annual conference on Computer graphics and interactive techniques - SIGGRAPH '00. ACM. pp. 145–156. doi:10.1145/344779.344855. ISBN 978-1581132083. Retrieved 2020-06-27.
  6. ↑ 6.0 6.1 "Inventing Entertainment: The Early Motion Pictures and Sound Recordings of the Edison Companies". Memory.loc.gov. w:Library of Congress. Retrieved 2020-12-09.
  7. ↑ 7.0 7.1 Blue, Logan; Warren, Kevin; Abdullah, Hadi; Gibson, Cassidy; Vargas, Luis; O’Dell, Jessica; Butler, Kevin; Traynor, Patrick (August 2022). "Detecting deep-fake audio through vocal tract reconstruction". Proceedings of the 31st USENIX Security Symposium: 2691–2708. ISBN 978-1-939133-31-1. Retrieved 2022-10-06.
  8. ↑ 8.0 8.1 Blue, Logan; Traynor, Patrick (2022-09-20). "Deepfake audio has a tell – researchers use fluid dynamics to spot artificial imposter voices". theconversation.com. w:The Conversation (website). Retrieved 2022-10-05. By estimating the anatomy responsible for creating the observed speech, it’s possible to identify the whether the audio was generated by a person or a computer.
  9. ↑ 9.0 9.1 Drew, Harwell (2020-04-16). "An artificial-intelligence first: Voice-mimicking software reportedly used in a major theft". w:washingtonpost.com. w:Washington Post. Retrieved 2019-07-22. Researchers at the cybersecurity firm Symantec said they have found at least three cases of executives’ voices being mimicked to swindle companies. Symantec declined to name the victim companies or say whether the Euler Hermes case was one of them, but it noted that the losses in one of the cases totaled millions of dollars.
  10. ↑ 10.0 10.1 Stupp, Catherine (2019-08-30). "Fraudsters Used AI to Mimic CEO's Voice in Unusual Cybercrime Case". w:wsj.com. w:The Wall Street Journal. Retrieved 2022-01-01.
  11. ↑ 11.0 11.1 Damiani, Jesse (2019-09-03). "A Voice Deepfake Was Used To Scam A CEO Out Of $243,000". w:Forbes.com. w:Forbes. Retrieved 2022-01-01. According to a new report in The Wall Street Journal, the CEO of an unnamed UK-based energy firm believed he was on the phone with his boss, the chief executive of firm’s the German parent company, when he followed the orders to immediately transfer €220,000 (approx. $243,000) to the bank account of a Hungarian supplier. In fact, the voice belonged to a fraudster using AI voice technology to spoof the German chief executive. Rüdiger Kirsch of Euler Hermes Group SA, the firm’s insurance company, shared the information with WSJ.
  12. ↑ "Fake voices 'help cyber-crooks steal cash'". w:bbc.com. w:BBC. 2019-07-08. Retrieved 2020-07-22.
  13. ↑ Franceschi-Bicchierai, Lorenzo (2020-07-23). "Listen to This Deepfake Audio Impersonating a CEO in Brazen Fraud Attempt". w:Vice.com. w:Vice (magazine). Retrieved 2022-01-03.
  14. ↑ 14.0 14.1 14.2 14.3 https://www.forbes.com/sites/thomasbrewster/2021/10/14/huge-bank-fraud-uses-deep-fake-voice-tech-to-steal-millions/
  15. ↑ Flatow, Ira (April 4, 2008). "1860 'Phonautograph' Is Earliest Known Recording". NPR. Retrieved 2012-12-09.
  16. ↑ https://www.theverge.com/2023/2/7/23587454/microsoft-bing-edge-chatgpt-ai
  17. ↑ https://www.reddit.com/r/MachineLearning/comments/gh9cbg/p_generate_handwriting_with_an_inbrowser/
  18. ↑ "What is IWR? (Intelligent Word Recognition)". eFileCabinet. 2016-01-04. Retrieved 2021-09-21.
  19. ↑ Edwards, Benj (2023-01-10). "Microsoft's new AI can simulate anyone's voice with 3 seconds of audio". w:Arstechnica.com. Arstechnica. Retrieved 2023-05-05. For the paper's conclusion, they write: "Since VALL-E could synthesize speech that maintains speaker identity, it may carry potential risks in misuse of the model, such as spoofing voice identification or impersonating a specific speaker. To mitigate such risks, it is possible to build a detection model to discriminate whether an audio clip was synthesized by VALL-E. We will also put Microsoft AI Principles into practice when further developing the models."
  20. ↑ 20.0 20.1 20.2 20.3 20.4 20.5 20.6 20.7 Authoritative up-to-date version of the Criminal Code chapter 20 On sexual offences can always be found at finlex.fi

    Translation to English by the Ministry of Justice: Criminal Code (39/1889) - Chapter 20 - Sexual offences (translation) as .pdf at oikeusministerio.fi (subject to possible revisions)
  21. ↑ "DALL·E 2". OpenAI. Retrieved 2023-04-22.
  22. ↑ Boháček, Matyáš; Farid, Hany (2022-06-14). "Protecting President Zelenskyy against Deep Fakes". arXiv:2206.12043 [cs.CV].
  23. ↑ Almutairi, Zaynab; Elgibreen, Hebah (2022-05-04). "A Review of Modern Audio Deepfake Detection Methods: Challenges and Future Directions". w:Algorithms (journal). doi:https://doi.org/10.3390/a15050155 Check |doi= value (help). Retrieved 2022-10-18.
  24. ↑ Rosner, Helen (2021-07-15). "A Haunting New Documentary About Anthony Bourdain". w:The New Yorker. Retrieved 2021-08-25. Check |author-link= value (help)
  25. ↑ https://www.partnershiponai.org/aiincidentdatabase/
  26. ↑ https://www.bbc.com/news/technology-55424730
  27. ↑ Johnson, R.J. (2019-12-30). "Here Are the New California Laws Going Into Effect in 2020". KFI. iHeartMedia. Retrieved 2021-01-23.
  28. ↑ "AB 602 - California Assembly Bill 2019-2020 Regular Session - Depiction of individual using digital or electronic technology: sexually explicit material: cause of action". openstates.org. openstates.org. Retrieved 2021-03-24.
  29. ↑ Mihalcik, Carrie (2019-10-04). "California laws seek to crack down on deepfakes in politics and porn". w:cnet.com. w:CNET. Retrieved 2021-01-23.
  30. ↑ "China seeks to root out fake news and deepfakes with new online content rules". w:Reuters.com. w:Reuters. 2019-11-29. Retrieved 2021-01-23.
  31. ↑ Statt, Nick (2019-11-29). "China makes it a criminal offense to publish deepfakes or fake news without disclosure". w:The Verge. Retrieved 2021-01-23.
  32. ↑ "Relating to the creation of a criminal offense for fabricating a deceptive video with intent to influence the outcome of an election". w:Texas. 2019-06-14. Retrieved 2021-01-23. In this section, "deep fake video" means a video, created with the intent to deceive, that appears to depict a real person performing an action that did not occur in reality
  33. ↑ https://capitol.texas.gov/BillLookup/History.aspx?LegSess=86R&Bill=SB751
  34. ↑ "New state laws go into effect July 1".
  35. ↑ 35.0 35.1 "§ 18.2-386.2. Unlawful dissemination or sale of images of another; penalty". w:Virginia. Retrieved 2021-01-23.
  36. ↑ "NVIDIA Open-Sources Hyper-Realistic Face Generator StyleGAN". Medium.com. 2019-02-09. Retrieved 2020-07-13.
  37. ↑ Harwell, Drew (2018-12-30). "Fake-porn videos are being weaponized to harass and humiliate women: 'Everybody is a potential target'". w:The Washington Post. Retrieved 2020-07-13. In September [of 2018], Google added “involuntary synthetic pornographic imagery” to its ban list
  38. ↑ Kuo, Lily (2018-11-09). "World's first AI news anchor unveiled in China". Retrieved 2020-07-13.
  39. ↑ Hamilton, Isobel Asher (2018-11-09). "China created what it claims is the first AI news anchor — watch it in action here". Retrieved 2020-07-13.
  40. ↑ Suwajanakorn, Supasorn; Seitz, Steven; Kemelmacher-Shlizerman, Ira (2017), Synthesizing Obama: Learning Lip Sync from Audio, University of Washington, retrieved 2020-07-13
  41. ↑ Giardina, Carolyn (2015-03-25). "'Furious 7' and How Peter Jackson's Weta Created Digital Paul Walker". The Hollywood Reporter. Retrieved 2020-07-13.
  42. ↑ ReForm - Hollywood's Creating Digital Clones (youtube). The Creators Project. 2020-07-13.
  43. ↑ Debevec, Paul. "Digital Ira SIGGRAPH 2013 Real-Time Live". Retrieved 2017-07-13.
  44. ↑ In this TED talk video at 00:04:59 you can see two clips, one with the real Emily shot with a real camera and one with a digital look-alike of Emily, shot with a simulation of a camera - Which is which is difficult to tell. Bruce Lawmen was scanned using USC light stage 6 in still position and also recorded running there on a w:treadmill. Many, many digital look-alikes of Bruce are seen running fluently and natural looking at the ending sequence of the TED talk video.
  45. ↑ Pighin, Frédéric. "Siggraph 2005 Digital Face Cloning Course Notes" (PDF). Retrieved 2020-06-26.
  46. ↑ https://ict.usc.edu/about/
  47. ↑ "Images de synthèse : palme de la longévité pour l'ombrage de Gouraud".
  48. ↑ Mechanismus der menschlichen Sprache nebst der Beschreibung seiner sprechenden Maschine ("Mechanism of the human speech with description of its speaking machine", J. B. Degen, Wien).
  49. ↑ History and Development of Speech Synthesis, Helsinki University of Technology, Retrieved on November 4, 2006
Retrieved from "https://stop-synthetic-filth.org/w/index.php?title=Synthetic_human-like_fakes&oldid=4754"