I’m working to try and get Convai Unreal Engine SDK 4.0.0-beta.19 with ACP voices and Claude 4.5 Opus to work as intended. I’ve followed the tutorial step by step. Experiencing the following issues:
Lipsync visemes are over-exaggerated and over-pronounced, even when using MetaHuman Blendshapes setting.
Lips intermittently fail to animate while audio plays correctly.
Emotion expressions appear over-pronounced.
Audio cutting out intermittently.
I note beta.18 changelog mentions command-line overrides for lipsync animation parameters. could you please clarify available parameters and recommended values for dampening amplitude? Also, where can emotion intensity be adjusted? They are over exaggerated.
It seems the SDK 4.0.0-beta.19 is not production ready and we need to stick with the SDK 3.6.8-beta.3 for now. At least it looks like the Digital Humans are actually speaking and cartoon expressions…
I presume you meant the above for the Audio cutting out intermittently bug. Yes, just tested on the 3.6.8 beta 3 project as well as 4.0.0 beta 19 project with Gemini 3.1 Flash Lite Preview and GPT-Realtime 1.5 and still the audio is intermittent, the lip stop working randomly. The long-term memory is not the best. Will continue testing! In the meantime, if we can fine tune the lip sync and emotion expression issues, make it accessible it would be wonderful to try and do a comparison and get it to a viable marketable product.