I've found that when I try out some of my competitors' voice assistants, they record a lot of phrases in advance, which they then reuse during the call. They do this to avoid making duplicate requests to elevenlabs (which gets very expensive very quickly when you want to scale +10,000 calls). Do you know how they do it? And would it be technically feasible to do it via VAPI? It's a very important point in terms of profitability if you want to run the bot on large volumes, and it would also save a bit of latency during the call.