Tag
This paper evaluates the recovery of clinically required content when patients interrupt clinical voice agents, testing various LLM configurations and finding that interruption robustness varies by context and model.