3,839
edits
Synthetic human-like fakes: Difference between revisions
Jump to navigation
Jump to search
m
Text replacement - "Laws against synthesis and related crimes" to "Laws against synthesis and other related crimes"
Juho Kunsola (talk | contribs) m (fmt) |
Juho Kunsola (talk | contribs) m (Text replacement - "Laws against synthesis and related crimes" to "Laws against synthesis and other related crimes") |
||
| (14 intermediate revisions by the same user not shown) | |||
| Line 1: | Line 1: | ||
'''Definitions''' | '''Definitions''' | ||
<section begin=definitions-of-synthetic-human-like-fakes /> | <section begin=definitions-of-synthetic-human-like-fakes /> | ||
When the '''[[Glossary#No camera|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-alikes|digital look-alike]]'''. | When the '''[[Glossary#No camera|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 '''[[Synthetic human-like fakes#Digital look-alikes|digital look-alike]]'''. | ||
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-alikes|digital sound-alike]]'''. | 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 '''[[Synthetic human-like fakes#Digital sound-alikes|digital sound-alike]]'''. | ||
::[[Synthetic human-like fakes|Read more about '''synthetic human-like fakes''']], see | ::[[Synthetic human-like fakes|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, [[Synthetic human-like fakes#Timeline of synthetic human-like fakes|examine the SSFWIKI '''timeline''' of synthetic human-like fakes]] or [[Mediatheque|view the '''Mediatheque''']]. | ||
<section end=definitions-of-synthetic-human-like-fakes /> | <section end=definitions-of-synthetic-human-like-fakes /> | ||
| Line 119: | Line 119: | ||
The Iframe below is transcluded from [https://google.github.io/tacotron/publications/speaker_adaptation/ ''''''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. | The Iframe below is transcluded from [https://google.github.io/tacotron/publications/speaker_adaptation/ ''''''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. | |||
{{#Widget:Iframe - Audio samples from Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis by Google Research}} | {{#Widget:Iframe - Audio samples from Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis by Google Research}} | ||
| Line 194: | Line 196: | ||
|access-date=2022-01-01 | |access-date=2022-01-01 | ||
|quote=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.}} | |quote=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.}} | ||
</ref> | |||
==== 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. ([https://soundcloud.com/jason-koebler/redacted-clip '''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 [https://www.nisos.com/blog/synthetic-audio-deepfake/ 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.<ref name="Vice reporting on 2020 digital sound-alike fraud attempt"> | |||
{{cite web | |||
|url=https://www.vice.com/en/article/pkyqvb/deepfake-audio-impersonating-ceo-fraud-attempt | |||
|title=Listen to This Deepfake Audio Impersonating a CEO in Brazen Fraud Attempt | |||
|last=Franceschi-Bicchierai | |||
|first=Lorenzo | |||
|date=2020-07-23 | |||
|website=[[w:Vice.com]] | |||
|publisher=[[w:Vice (magazine)]] | |||
|access-date=2022-01-03 | |||
|quote=}} | |||
</ref> | </ref> | ||
| Line 257: | Line 276: | ||
[[File:Spectrogram-19thC.png|thumb|right|640px|A [[w:spectrogram]] of a male voice saying 'nineteenth century']] | [[File:Spectrogram-19thC.png|thumb|right|640px|A [[w:spectrogram]] of a male voice saying 'nineteenth century']] | ||
== Text syntheses == | == Text syntheses == | ||
| Line 334: | Line 340: | ||
* [https://github.com/topics/handwriting-recognition GitHub topic '''handwriting-recognition'''] contains 238 repositories as of September 2021. | * [https://github.com/topics/handwriting-recognition GitHub topic '''handwriting-recognition'''] contains 238 repositories as of September 2021. | ||
== 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. | |||
* [https://arxiv.org/abs/1910.11690 ''''''Fast and High-Quality Singing Voice Synthesis System based on Convolutional Neural Networks'''''' at arxiv.org], a 2019 singing voice synthesis technique using [[w:convolutional neural network|w:convolutional neural networks (CNN)]]. Accepted into the 2020 [[w:International Conference on Acoustics, Speech, and Signal Processing|International Conference on Acoustics, Speech, and Signal Processing (ICASSP)]]. | |||
* [http://compmus.ime.usp.br/sbcm/2019/papers/sbcm-2019-7.pdf ''''''State of art of real-time singing voice synthesis'''''' at compmus.ime.usp.br] presented at the 2019 [http://compmus.ime.usp.br/sbcm/2019/program/ 17th Brazilian Symposium on Computer Music] | |||
* [http://theses.fr/2017PA066511 ''''''Synthesis and expressive transformation of singing voice'''''' at theses.fr] [https://www.theses.fr/2017PA066511.pdf as .pdf] a 2017 doctorate thesis by [http://theses.fr/227185943 Luc Ardaillon] | |||
* [http://mtg.upf.edu/node/512 ''''''Synthesis of the Singing Voice by Performance Sampling and Spectral Models'''''' at mtg.upf.edu], a 2007 journal article in the [[w:IEEE Signal Processing Society]]'s Signal Processing Magazine | |||
* [https://www.researchgate.net/publication/4295714_Speech-to-Singing_Synthesis_Converting_Speaking_Voices_to_Singing_Voices_by_Controlling_Acoustic_Features_Unique_to_Singing_Voices ''''''Speech-to-Singing Synthesis: Converting Speaking Voices to Singing Voices by Controlling Acoustic Features Unique to Singing Voices'''''' at researchgate.net], a November 2007 paper published in the IEEE conference on Applications of Signal Processing to Audio and Acoustics | |||
* [[w:Category:Singing software synthesizers]] | |||
---- | |||
= Timeline of synthetic human-like fakes = | = Timeline of synthetic human-like fakes = | ||
| Line 339: | Line 360: | ||
== 2020's synthetic human-like fakes == | == 2020's synthetic human-like fakes == | ||
[[File:Appearance of Queen Elizabeth II stolen by Channel 4 in Dec 2020 (screenshot at 191s).png|thumb|right|480px|In Dec 2020 Channel 4 aired a Queen-like fake i.e. they had thieved the appearance of Queen Elizabeth II using deepfake methods.]] | |||
* '''2021''' | Science and demonstration | In the NeurIPS 2021 held virtually in December researchers from Nvidia and [[w:Aalto University]] present their paper [https://nvlabs.github.io/stylegan3/ '''''Alias-Free Generative Adversarial Networks (StyleGAN3)''''' at nvlabs.github.io] and associated [https://github.com/NVlabs/stylegan3 implementation] in [[w:PyTorch]] and the results are deceivingly human-like in appearance. [https://nvlabs-fi-cdn.nvidia.com/stylegan3/stylegan3-paper.pdf StyleGAN3 paper as .pdf at nvlabs-fi-cdn.nvidia.com] | * '''2021''' | Science and demonstration | In the NeurIPS 2021 held virtually in December researchers from Nvidia and [[w:Aalto University]] present their paper [https://nvlabs.github.io/stylegan3/ '''''Alias-Free Generative Adversarial Networks (StyleGAN3)''''' at nvlabs.github.io] and associated [https://github.com/NVlabs/stylegan3 implementation] in [[w:PyTorch]] and the results are deceivingly human-like in appearance. [https://nvlabs-fi-cdn.nvidia.com/stylegan3/stylegan3-paper.pdf StyleGAN3 paper as .pdf at nvlabs-fi-cdn.nvidia.com] | ||
| Line 366: | Line 387: | ||
* '''2021''' | '''<font color="red">crime / fraud</font>''' | {{#lst:Synthetic human-like fakes|2021 digital sound-alike enabled fraud}} | * '''2021''' | '''<font color="red">crime / fraud</font>''' | {{#lst:Synthetic human-like fakes|2021 digital sound-alike enabled fraud}} | ||
* '''<font color="green">2020</font>''' | '''<font color="green">counter-measure</font>''' | | * '''<font color="green">2020</font>''' | '''<font color="green">counter-measure</font>''' | The [https://incidentdatabase.ai/ ''''''AI Incident Database'''''' at incidentdatabase.ai] was introduced on 2020-11-18 by the [[w:Partnership on AI]].<ref name="PartnershipOnAI2020">https://www.partnershiponai.org/aiincidentdatabase/</ref> | ||
* '''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.<ref name="Queen-like deepfake 2020 BBC reporting">https://www.bbc.com/news/technology-55424730</ref> [https://www.youtube.com/watch?v=IvY-Abd2FfM&t=3s View in YouTube] | |||
* '''2020''' | reporting | [https://www.wired.co.uk/article/deepfake-porn-websites-videos-law "''Deepfake porn is now mainstream. And major sites are cashing in''" at wired.co.uk] by Matt Burgess. Published August 2020. | * '''2020''' | reporting | [https://www.wired.co.uk/article/deepfake-porn-websites-videos-law "''Deepfake porn is now mainstream. And major sites are cashing in''" at wired.co.uk] by Matt Burgess. Published August 2020. | ||
| Line 373: | Line 396: | ||
** [https://www.cnet.com/news/mit-releases-deepfake-video-of-nixon-announcing-nasa-apollo-11-disaster/ Cnet.com July 2020 reporting ''MIT releases deepfake video of 'Nixon' announcing NASA Apollo 11 disaster''] | ** [https://www.cnet.com/news/mit-releases-deepfake-video-of-nixon-announcing-nasa-apollo-11-disaster/ Cnet.com July 2020 reporting ''MIT releases deepfake video of 'Nixon' announcing NASA Apollo 11 disaster''] | ||
* '''2020''' | US state law | {{#lst: | * '''2020''' | US state law | {{#lst:Laws against synthesis and other related crimes|California2020}} | ||
* '''2020''' | Chinese legislation | {{#lst: | * '''2020''' | Chinese legislation | {{#lst:Laws against synthesis and other related crimes|China2020}} | ||
== 2010's synthetic human-like fakes == | == 2010's synthetic human-like fakes == | ||
| Line 380: | Line 403: | ||
** Reporting [https://www.technologyreview.com/2020/08/28/1007746/ai-deepfakes-memes/ '''''Memers are making deepfakes, and things are getting weird''''' at technologyreview.com], 2020-08-28 by Karen Hao. | ** Reporting [https://www.technologyreview.com/2020/08/28/1007746/ai-deepfakes-memes/ '''''Memers are making deepfakes, and things are getting weird''''' at technologyreview.com], 2020-08-28 by Karen Hao. | ||
* '''2019''' | demonstration | In September 2019 [[w:Yle]], the Finnish [[w:public broadcasting company]], aired a result of experimental [[w:journalism]], [https://yle.fi/uutiset/3-10955498 '''a deepfake of the President in office'''] [[w:Sauli Niinistö]] in its main news broadcast for the purpose of highlighting the advancing disinformation technology and problems that arise from it. | * '''2019''' | demonstration | In September 2019 [[w:Yle]], the Finnish [[w:public broadcasting company]], aired a result of experimental [[w:journalism]], [https://yle.fi/uutiset/3-10955498 '''a deepfake of the President in office'''] [[w:Sauli Niinistö]] in its main news broadcast for the purpose of highlighting the advancing disinformation technology and problems that arise from it. | ||
* '''2019''' | US state law | {{#lst: | * '''2019''' | US state law | {{#lst:Laws against synthesis and other related crimes|Texas2019}} | ||
* '''2019''' | US state law | {{#lst: | * '''2019''' | US state law | {{#lst:Laws against synthesis and other related crimes|Virginia2019}} | ||
* '''2019''' | Science | [https://arxiv.org/pdf/1809.10460.pdf '''''Sample Efficient Adaptive Text-to-Speech''''' .pdf at arxiv.org], a 2019 paper from Google researchers, published as a conference paper at [[w:International Conference on Learning Representations]] (ICLR)<ref group="1st seen in" name="ConnectedPapers suggestion on Google Transfer learning 2018"> https://www.connectedpapers.com/main/8fc09dfcff78ac9057ff0834a83d23eb38ca198a/Transfer-Learning-from-Speaker-Verification-to-Multispeaker-TextToSpeech-Synthesis/graph</ref> | * '''2019''' | Science | [https://arxiv.org/pdf/1809.10460.pdf '''''Sample Efficient Adaptive Text-to-Speech''''' .pdf at arxiv.org], a 2019 paper from Google researchers, published as a conference paper at [[w:International Conference on Learning Representations]] (ICLR)<ref group="1st seen in" name="ConnectedPapers suggestion on Google Transfer learning 2018"> https://www.connectedpapers.com/main/8fc09dfcff78ac9057ff0834a83d23eb38ca198a/Transfer-Learning-from-Speaker-Verification-to-Multispeaker-TextToSpeech-Synthesis/graph</ref> | ||