88Z AI, 28 September — NetEase Youdao has open-sourced two real-time interaction models, R2T2 and T3PO, from its Ziya Live series. Optimized for streaming speech recognition and translation respectively, they quickly topped relevant trending lists on Hugging Face.
At a glance
- R2T2 and T3PO topped Hugging Face trending lists due to low latency and high quality.
- The models are compatible with llama.cpp, Ollama, and GGUF quantization for diverse deployment options.
The update
NetEase Youdao has officially open-sourced two real-time models from its Ziya Live series: Confucius4-R2T2 for streaming speech recognition and Confucius4-T3PO for streaming translation. Leveraging "stable incremental" technology, these models enable efficient interactions by recognizing and translating speech in real time. R2T2 maintains an average latency of 200 to 600 milliseconds, while T3PO balances low latency with high translation quality. Following their release, both models quickly topped the ASR and Translation trending lists on Hugging Face due to strong community interest. They now support ZeroGPU Demo, have been adapted for audio.cpp, undergone GGUF quantization, and achieved compatibility with llama.cpp and Ollama, significantly lowering the barrier for developers.
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.