Do companies use machine learning for face animation at this point? E.g. capture a bunch of facial mocap data and phoneme streams and train a transformer model on how the face would move given a new stream? I guess it's probably easier just to translate phonemes to facial movements with much the same outcome?