88Z AI, 25 June — NetEase Youdao has officially open-sourced the Confucius4-TTS speech synthesis model. The model achieves breakthroughs in zero-shot cloning, multi-language accent-free synthesis, and emotion transfer, utilizing an end-to-end architecture that supports local offline deployment.
At a glance
- Supports 3-second zero-shot cloning and accent-free synthesis in 14 languages
- Uses an end-to-end architecture with 54GB weights for local offline deployment
The update
The newly open-sourced Confucius4-TTS model from NetEase Youdao achieves three core breakthroughs in speech synthesis technology. First, it features robust zero-shot voice cloning, capable of replicating a voice with just three seconds of audio. Second, it supports cross-language synthesis in 14 languages while maintaining a natural, accent-free sound. Additionally, the model can accurately transfer emotion and prosody. Architecturally, it integrates a speech encoder, a large language model, and flow matching generation to build an efficient end-to-end system. With a full weight size of 54GB, it fully supports local offline deployment, meeting the needs of applications requiring high data privacy and low latency.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 网易有道Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.