A production voice agent has a latency budget of roughly 800 milliseconds before a conversation starts to feel slow. That budget is spent across eight sequential stages, typically split across three to five different vendors depending on how the stack is assembled. This post maps that voice
