Visemes and Phonics go together. Variations in speech, like dialect and origin do have an impact, but there are more similarities than differences.
Visemes are about the shape and Phonics is about the sound.
Interesting when people try and match the visemes they start making the sound and looking at there lips... which sounds obvious, but we exaggerate both the sound and the shape. It seems then to get CA to work anywhere near, we have to exaggerate, but this isn't normal and this seems to lead to a missmatch.
Is it possible to try and relate the visemes to the phonetical sounds. With the visemes available in CA, you have to seconded guess, and tryout the related shapes in an attempt to match the sound. With a 5 minute audio this can be a lot of work trying to and correct. The auto sync is a great idea, but it doesn't do that much of a good job.
As a help, could there be a printout that relates visemes to phonics?
I've notice that people are now using cycles more when producing speech shapes are there are some amazing results. They realise that the sound produced is rarely just one shape.
This must take a fair bit of work, so I was wondering if it would be possible to save each cycle, related to it's particular sound and have this stored as a library which can be accessed accross projects. This could be accessed like a Properties, History, or control panel.
Speech is the first step in the recording process and I think CA's recording system works ok, it is only the mismatch of the spoken word and the Viseme it uses that is the problem. Maybe a few more visemes are needed and the algorithm needs a little TLC?