[photonemitter]

This reminded me; People seem way too comfortable / in the dark when it comes to voice-cloning, which, with the added data of general speech-patterns, is likely to become a major issue wrt political/legal issues.

see: https://github.com/CorentinJ/Real-Time-Voice-Cloning (think it made its rounds here on HN before)

Conceptually, even though it currently is a market-strategy where the benefits are as you point out. In the longer term this could be argued as an uncomfortable seat of power to whoever sits on the user-data. (without bringing up too much of that issue here... might be time to generate obfuscation of voice similarly to the obfuscation of google-searching which I've seen on here earlier=)