Nearly two years ago I was at the We Are Developers conference in Berlin; one of the talks was by someone who cloned their own voice, connected it to a messenger (IIRC WhatsApp but it doesn't matter), and had it listen to voice messages, pass through to an LLM prompted to answer in his style, and synthesise a voice answer.

Apparently it worked too well, as it was replying in less than the time it would take a human to listen to the input message. And it fooled people who knew him.