Earlier in the week we noticed from some customer feedback that there were some issues with visemes and voices.
Response time of the model was longer than expected - and this negatively affected customer experience. We're doing a post mortem internally and will fix this going forward.