When I tested with only creating a few morphs, the speech was realistic but as it got more complex the mouth movement starts vibrating around the mouth area.

I've been observing the evolution of Visemes for years and their associated thumbnails to help identify to users what the face expression is supposed to be.

While it's getting better, there are still incorrect thumbnails and unnecessary visemes that lessor amount could already work with other visemes to produce the same sound and not create intermediate duplicates with slight offsets.

For example AH and AE - the only difference in the image to produce the sound is tongue position, the mouth shape should be the same.

Whether its a CC character or custom it doesn't matter, the face will look like it vibrates in a full sentence on acting animations such as Talk_M.

Another example ER and R , thumbs are different because you assume they are different sounds but the Viseme AE already helps make ER (like AE+R = Air or EE + R = ear)

So my suggestion is to completely remove ER and change the thumbnails of AH and AE to be the same picture but the tongue in AH is down and AE the tongue in the back is up.

This is my observation and experience, I hope you experiment a little more and try these modifications.

I may make a video later showing the results.