Zero-shot open-source TTS from the Amphion toolkit. Non-autoregressive masked generative model with state-of-the-art voice cloning.