Step-Audio-TTS-3B represents the industry's first Text-to-Speech (TTS) model trained on a large-scale synthetic dataset utilizing the LLM-Chat paradigm