ai-agents-conversational-tts-i-need-to-add-natural-speech-output-to-an-llm-based--fe9b6f2e

THEN Use ChatTTS in an LLM pipeline: after receiving text from the LLM, call `chat.infer()` with the text and optional speaker embedding from `chat.sample_random_speaker()`. For dialogue, maintain speaker embeddings per turn. The model supports mixed Chinese/English and controls like laughter. Use streaming audio generation if available (see roadmap).

```python
# Paraphrase: Integration with LLM
chat = ChatTTS.Chat()
chat.load(compile=False)
speaker = chat.sample_random_speaker()  # fix timbre

while True:
    user_input = input()
    llm_response = llm.generate(user_input)
    wav = chat.infer([llm_response], params_infer_code=ChatTTS.Chat.InferCodeParams(spk_emb=speaker))
    play_audio(wav)
```