That's correct, however the script from the LLM is converted into speech via the TTS. And there is no way to accurately control the speed of the generated voice at this moment, atleast if using 11Labs.