This article explores the evolution of speech synthesis technology, from the limitations of traditional TTS models to the integration of large language models, analyzing the technical principles of audio encoders and neural codecs, and how modern TTS models achieve context-aware conversational speech synthesis.