Are you using something like Pipecat or the VAPI API? How are you handling your voice->voice pipeline and your audio infra?