Hacking voice using ML, AKA DeepVoice
As with the examples of imagery for video and fingerprints, ML applications are also suited to manipulate voice content and even to generate realistic fake voices from nothing. The original impetus for most voice generation and cloning was supposed to be for voice commands and voice assistant technology applications.
Because humans typically prefer to feel that they are interacting with other humans, retailers have spent significant time and money building out realistic human voice assistants and online support solutions. Those bots and fake human voices are composed of either real past human recordings that have been strung together to reproduce realistic human voice outputs, or they are built using ML backend systems that literally generate human voices out of thin air.
When Siri, Alexa, or our GPS use voice generation, it is usually quickly obvious that it is a bot or non-human speaking to the user. This is because virtually every legacy...
