Bethesda doing is semi-realistic face similar what Square dose with Final Fantasy...the difference is SE does MUCH better job.
While I agree and playing Cyberpunk and FF7R right after playing 130hrs of Starfield was like AH, ACTUAL PEOPLE there is one thing to be said to be completely fair: Bethesda uses something like text-to-speech except it is text-to-face to animate them, then they add emotion queues. This requires all faces to fit within certain parameters by which the software functions and it manipulates them indescrinimately of their particular features. Something about this apparently helps them have as many talking NPCs as they do while keeping up with many localizations.
That doesn't mean the end results aren't way below doing it in more traditional ways, and one can still say they don't think it is worth it and that they should invest in what it takes to do things manually or with mocap to bring it up to par with other AAA budget games, but I'm just saying there is a different reason for it than them trying and miserably failing to do it the traditional ways. It is a system they have kept since Oblivion, so however they have tried to improve it, it is no doubt massively outdated in basic concept and function.