#zero-shot TTS

Step-Audio-EditX is an open-source 3B audio LLM that edits speech at the token level, enabling iterative, controllable changes to emotion, style, and paralinguistic cues through large-margin synthetic data and token-level PPO.

'VoXtream is an open-source full-stream zero-shot TTS that begins speaking from the first word, delivering sub-120 ms first-packet latency with compiled GPUs and continuous 80 ms audio frames.'