Hi,
We are currently using ver 1.7.3.1 of FaceFX to create a large amount of auto-generated dialog, along with text analysis and text markups for emoting etc. Our face shapes are generated using morph targets, as we only have a couple of faces on screen to deal with at any time.
In regards to the auto-analysis FaceFX does using the phonemes for lipsync - I noticed that for speech which is said quickly, you tend to get tight clusters of phonemes being used which can cause the mouth shapes blur into each other and not be clearly defined.
For example I have the text... "look it was something like an", FaceFX returns the result... Look = L UH K, it = IH T, was = W AH Z, something = S AH M P TH IH NG, like = L AY K, an = AH N. While this looks correct in the mapping, the speed at which it is said in the voice track causes the analysis to compress some of the phoneme shapes into thin blue boxes in the animation viewer and consequently the shapes are rapidly passed through during animation which can look too busy and unclear.
Is there some way to control having some phonemes omitted during the analysis, such that you end up with a cleaner result?
Should we just keep messing with text itself in order to find an analysis that works?
Unfortunately that's just a fact of life in the 1.x versions.
If you upgrade to 2010 there is a phoneme funneling Python script that takes care of this and in the about to be released 2012 that has been improved and built into the application itself. The process removes phonemes that "don't matter", especially in fast speech.
Thanks for info. I figured we wouldn't get those features using the old version, just wondering if there was something I missed. Unfortunately we're not likely to update for our current project.
In the old version if you come across a particularly bad case you can always just delete phonemes in the phoneme bar. The newer versions also have a curve smoothing feature to smooth things out.
Remember you can use data from the latest version of FaceFX in 1.7.3.1 by using the XML pipeline. You don't have to "fully" upgrade to FaceFX 2012.