For all my assistants I’m using eleven labs voices and assembly transcription. I make sure to speed up latency to the max for voice. Oh yeah I also use the latest 4o mini (not real-time) Hopefully this configuration helps out