Adjusting mouth shape to better match phonemes (Blendshape solve options?)

AI & Data Science Digital Humans (closed) Audio2Face (closed)
1

Hello! Audio2Face is wonderful! Thank you for all the hard work!

In one of the NVIDIA video tutorials (Animating MetaHuman with Omniverse Audio2Face and Autodesk Maya - YouTube) I saw that the blendshape solver options were used to improve mouth shapes.

I’m having trouble with getting good mouth/lip shapes to match M, P, B. Are there any good tips to try to get better lip movement?

2

Hi @andrew199 thanks for your interest in Audio2Face.

Making good M, P, B shapes are sometimes difficult depending on the emotion states. We’re working towards providing a better deep learning network in future releases.
In the meantime, the general tips would be like this

  • As in the video, use some symmetry constraints if the lip shape is not symmetric.
  • If lip is not closing properly, try using lipOpenOffset in the post-process to close the lips. (attached image)
  • Neutral emotion tends to give good MBP poses, so use strong neutral emotion and a bit of other emotions.

image

3

@yseol thanks for the tips! I will try them out

4

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.