Balancing ultra-low latency and cloud economics in virtual voice agents