I mostly being using the AWS bedrock LLM, TTS and STT. I have noticed in the terminal that only after you get all the chunks from LLM, TTS starts to work. When the answer are long the latency tend to be really high. Is this the intended way? Or are there any ways to optimise for latency.