Abstract: This paper presents a streaming text-to-speech (TTS) framework for real-time speech synthesis in LLM-driven conversational systems. We extend FastSpeech2, a non-autoregressive model, with ...