Bear algorithm to derive speaker-dependent visemes. Input is a confusion matrix where labels are max 3 characters.
If you use this algorithm please reference the source;
@article{bear2017phoneme, title={Phoneme-to-viseme mappings: the good, the bad, and the ugly}, author={Bear, Helen L and Harvey, Richard}, journal={Speech Communication}, volume={95}, pages={40--67}, year={2017}, publisher={Elsevier} }