One could argue that the audio used to train the deepfakes was the actual infringement, as it is essentially a copy of presumably copyrighted audio, even if not

One could argue that the audio used to train the deepfakes was the actual infringement, as it is essentially a copy of presumably copyrighted audio, even if not republished in its original form. This would be similar to the Google Books scanning issue. Their original plan was to scan all the books in the world so users could search them. Though Google did provide extended samples in the search results, the core issue was the scanning, which was literally a copy of the books into Google's database without permission. This was considered infringement. [1] Where the original sound comes from and who owns the copyright for that, I don't know. Maybe the person used CNN videos, so Time Warner should be involved as well. 1. https://www.wikipedia.org/wiki/Authors_Guild,_Inc._v._Google,_Inc https://www.wikipedia.org/wiki/Authors_Guild,_Inc._v._Google....