The data to be translated: The Webmaster Home reports on an innovative speech synthesis system named NaturalSpeech 3, which utilizes decomposition codecs and diffusion models to generate natural speech in zero-shot scenarios. This system achieves fine modeling of speech waveforms through neural codecs, outperforming existing TTS systems in multiple benchmark tests. Researchers propose to enhance synthetic speech detection models to address potential abuse risks, in line with Microsoft's principles of responsible AI.