3,839
edits
Main Page: Difference between revisions
updating text + link to the transclusion of 'Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis'
Juho Kunsola (talk | contribs) m (commas) |
Juho Kunsola (talk | contribs) (updating text + link to the transclusion of 'Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis') |
||
| Line 28: | Line 28: | ||
Since the early 00's it has become (nearly) impossible to determine in still or moving pictures what is an image of a human, imaged with a (movie) camera and what on the other hand is a simulation of an image of a human imaged with a simulation of a camera. When there is '''no camera''' and the target being imaged with a simulation looks deceptively like some real human, dead or living, it is a '''[[digital look-alikes|digital look-alike]]'''. | Since the early 00's it has become (nearly) impossible to determine in still or moving pictures what is an image of a human, imaged with a (movie) camera and what on the other hand is a simulation of an image of a human imaged with a simulation of a camera. When there is '''no camera''' and the target being imaged with a simulation looks deceptively like some real human, dead or living, it is a '''[[digital look-alikes|digital look-alike]]'''. | ||
Now in the 2010's the equivalent thing is happening to our '''voices''' i.e. they '''can be stolen''' to some extent with the 2016 prototypes like [[w:Adobe Inc.]]'s [[w:Adobe Voco|w:Adobe Voco]] and [[w:Google]]'s [[w:DeepMind]] [[w:WaveNet]] and '''made to say anything'''. When it is not possible to determine with human testing or testing with technological means what is a recording of some living or dead person's real voice and what is a simulation it is a '''[[digital sound-alikes|digital sound-alike]]'''. | Now in the late 2010's the equivalent thing is happening to our '''voices''' i.e. they '''can be stolen''' to some extent with the 2016 prototypes like [[w:Adobe Inc.]]'s [[w:Adobe Voco|w:Adobe Voco]] and [[w:Google]]'s [[w:DeepMind]] [[w:WaveNet]] and '''made to say anything'''. When it is not possible to determine with human testing or testing with technological means what is a recording of some living or dead person's real voice and what is a simulation it is a '''[[digital sound-alikes|digital sound-alike]]'''. 2018 saw the publication of Google Research's sound-like-anyone machine [[#'Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis' 2018 by Google Research (external transclusion)|(transcluded below)]] at the [[w:NeurIPS]] conference and by the end of '''2019''' Symantec research had learned of 3 cases where digital sound-alike technology '''had been used for crimes'''.<ref name="WaPo2019"> | ||
https://www.washingtonpost.com/technology/2019/09/04/an-artificial-intelligence-first-voice-mimicking-software-reportedly-used-major-theft/</ref> | https://www.washingtonpost.com/technology/2019/09/04/an-artificial-intelligence-first-voice-mimicking-software-reportedly-used-major-theft/</ref> | ||