GitHub radar
Youdao's model transcribes speech as you speak
NetEase Youdao released a streaming speech recognition model with under a second of delay — free to install, for transcribing audio almost live.
Confucius4-R2T2 is described by its authors as a low-latency and high-accuracy true streaming Automatic Speech Recognition model. It was built by the team at NetEase Youdao, a Chinese dictionary and translation service. The model listens to speech in short chunks, from eighty milliseconds up to two seconds, and outputs text right away, without waiting for the end of a phrase and without rewriting what it already said. The average latency is 200 to 600 milliseconds, and the authors say the accuracy is close to offline, non-streaming recognition. It is built on top of Qwen3-ASR, a 1.7 billion parameter model from Alibaba, fine-tuned by the Youdao team. Besides Chinese and English it also recognizes about a dozen other languages, including Russian. The model itself has about 2 billion parameters, which is small enough to run on a regular laptop.
Why a vibe-coder should care
For our content pipeline this is about subtitles and live captions: it produces text with under a second of delay, so it fits live streams or a quick transcript of a video. The strong side is speed, most streaming models cannot hold that kind of latency. The weak side is that the model is tuned mainly for Chinese and English, so for Russian the accuracy is not guaranteed to the same degree. You can install it for free, without anyone else's API key, but it needs Docker and some setup, it is not a one-click application.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up the speech recognition model Confucius4-R2T2 by NetEase Youdao for me: https://huggingface.co/netease-youdao/Confucius4-R2T2 — deploy it locally following the card's instructions (via Docker) and show me how to send it audio and get text back almost in real time.
A regular laptop with 8 GB of RAM or more — the model itself takes about 4 GB.
Open on Hugging Face▌ More finds