Phone ordering assistant for a pizza chain

A pizza chain wants an assistant that takes orders over the phone. The caller speaks, and the assistant answers out loud. It asks about sizes and toppings, reads the order back and sends it to the store. On Friday evenings about 1,500 calls run at once across 400 stores.

Each turn uses three hosted services. A speech recognizer turns the caller's words into text, and it knows the caller has finished about 300 ms after they go quiet. A language model writes the reply. Its first word comes about 600 ms after the request, then about 80 words a second follow. Replies run 15 to 40 words. A speech service turns the reply into audio, and its first audio is ready about 150 ms after it gets the text.

Callers talk over the assistant or hang up when a reply takes more than about 1 second to start.

Lay out one turn from the caller's last word to the assistant's first word, with the time each step takes.