You could prompt the LLM to add pauses in its responses as per 11Labs requested format and it might work. Haven't tried that through Vapi yet so cannot confirm but this is the only solution I can think of.