Here's my first proper test of using cleaned motion capture data and facial retargeting in Blender. Voice acting is still based on the on-set recording; the audio will be dubbed later.
Overall, I think the expressions work pretty nicely in sync. Rendered using Blender's native viewport.