If you are live-streaming, are you using obs or similar? Such streaming software often introduces lag in video vs audio which you need to adjust in OBS.
The other thing to try is to do a recording in CH then compare the live visemes with "compute from lipsync". Are they the same? Is the recorded audio quality (and hence your live audio) good quality? If audio quality that ch records is good and the two are different, then how fast is your machine? Is it struggling to keep up with the cpu to do the puppet animation and lipsync audio processing at the same time? The software has to best guess if given a time limit