How would you handle voice synthesis trade-offs between latency and quality for Assort Health's voice AI agents?
