“The end-to-end turn latency for OpenAI's GPT-4o voice mode can be as low as 300-320 milliseconds, but typically averages around 600-700 milliseconds.”