Vapi (in this context) has nothing to do with how the voice outputs the text - it simply reads what ChatGPT spits back at it from the API, and ElevenLabs handles the streaming and output of the voice from the model.