Guys I am seeing huge difference between latency on the VAPI dashboard and when I use it on my web client. Initializing takes 5-6 seconds which is too much. Any tips on how can I reduce that? PS : Using the smallest models for fastest results, using deepgram for STT.