What the system should do is if a detects an angry tone, it repeats the customers voice back to them with a slight delay. It is basically impossible to talk like that. It's like a bad phone call. Then once the person stops talking for a bit, it goes back to normal.... I now want to make this system.
I work in the telecom industry in a call center environment. I just saw an addon for the platform we use with this exact capability. I laughed, rolled my eyes & shared it with peers for them to laugh.
I would personally never recommend implementing something that can change the voice of the caller. Without having researched the product at all I will just say I cannot recommend any company install this capability. My primary reason for concern would be if legal proceedings have to go forward, what would stop either side from saying it was AI generated voice and not the actual caller? (Yes, for all I know the call recording may be stored raw.)
I'd also like to point out that this makes the mechanical "customer service can't help people" problem worse. They'll be all friendly while you're mad as hell. It makes the whole interaction even less genuine. The customer is probably angry because there isn't a human conversation to be had to begin with. I don't know why call center people even put up with that from their own organizations