The technology is getting impressive! The voices no longer sound robotic, like we're used to hearing text-to-speech. But they do sound awkward, in an uncanny valley way; something is just off about them.
But there is one cool aspect of this technology that this article doesn't mention: Text-to-speech can be applied to texts created by a GPT-like algorithm at run time, something that is completely impossible to do with traditional voice acting. That way the commentators can react to emergent gameplay that would be basically impossible to predict ahead of time. And they can mention players by name, and players could even type in any name they can think of for their teams and hear the announcers talk about it.