It's because oai are going for low latency. I've got pause free working locally by running the tts model after the llm has finished generating a huge chunk of the response.

And fine tuning on 15 minutes of the voice you're targeting makes it sound natural

Maybe because the OpenAI App where you talk to it is responding in Realtime.

But the NotebookLM is like a recorded pod-cast, so it can think about it before laying down a track?