I think one of the major medium-term effects of AI voice acting is going to be a slump in quality of product delivered to the public. AI output quality very rarely exceeds "serviceable, probably a bit wonky" in quality, and would take way more time and resources to try and iron out the kinks of a particular output, so the significant savings for the developer/producer come from just spamming the generate button a few times until they get one that they feel is "good enough". So expect a return to the era of off/flat readings that was caused by VAs being given context-less lists of lines, often not in narrative order, with little if any direction. It's cheap and the suits will convince themselves that audiences won't care enough to not consume product.