Add library name and paper link

#5
by nielsr HF Staff - opened
Files changed (1) hide show
  1. README.md +9 -6
README.md CHANGED
@@ -1,4 +1,10 @@
1
  ---
 
 
 
 
 
 
2
  tags:
3
  - confucius4
4
  - r2t2
@@ -9,11 +15,6 @@ tags:
9
  - speech-recognition
10
  - vllm
11
  - multilingual
12
- base_model: Qwen/Qwen3-ASR-1.7B
13
- pipeline_tag: automatic-speech-recognition
14
- license: other
15
- license_name: netease-model-use-license-agreement
16
- license_link: https://raw.githubusercontent.com/netease-youdao/Confucius4-R2T2/refs/heads/master/MODEL_LICENSE
17
  ---
18
 
19
  <div align="center">
@@ -48,6 +49,8 @@ license_link: https://raw.githubusercontent.com/netease-youdao/Confucius4-R2T2/r
48
  </div>
49
  <br>
50
 
 
 
51
  Confucius4-R2T2 is a low-latency and high-accuracy true streaming Automatic Speech Recognition (ASR) model that features fine-grained and configurable decoding chunks from 80 ms to 2 s. The model operates in append-only output mode: committing transcript text permanently without revising previous words, which is critical for applications where text must be processed or acted upon instantly. This results in a smoother user experience, avoiding disruptive text revisions and visual flickering in real-time applications, such as Real-Time Live Captioning & Subtitling, Downstream NLP Pipelines & LLM Agents, Simultaneous Speech Translation, etc.
52
 
53
  R2T2, short for Real Real-Time Transcription, is built upon the Qwen3-ASR model. And it is trained with a unique set of data construction techniques including stable-prefix data, forced time-alignment data, and token-level audio segmentation. Combined with a Longest Stable Prefix (LSP) learning paradigm (tech report will be released soon), R2T2 can dynamically determine when a stable prefix can be safely emitted and when additional audio context is needed. By exposing only stable prefixes, the model provides high-quality context that conditions subsequent predictions while guaranteeing that previously emitted text remains unchanged. Despite its streaming design, R2T2 maintains strong accuracy in offline recognition.
@@ -724,4 +727,4 @@ If you use this repository or the R2T2 checkpoint in your research, please cite
724
  R2T2 uses **dual licensing** to distinguish the source code from the model weights:
725
 
726
  - **Code** in the accompanying GitHub repository is released under the [Apache License 2.0](https://github.com/netease-youdao/Confucius4-R2T2/blob/master/LICENSE) and is free to use, modify, and redistribute (including commercially) under the terms of that license.
727
- - **Model weights** are released under the [NetEase Model Use License Agreement](https://github.com/netease-youdao/Confucius4-R2T2/blob/master/MODEL_LICENSE).
 
1
  ---
2
+ base_model: Qwen/Qwen3-ASR-1.7B
3
+ license: other
4
+ license_name: netease-model-use-license-agreement
5
+ license_link: https://raw.githubusercontent.com/netease-youdao/Confucius4-R2T2/refs/heads/master/MODEL_LICENSE
6
+ pipeline_tag: automatic-speech-recognition
7
+ library_name: transformers
8
  tags:
9
  - confucius4
10
  - r2t2
 
15
  - speech-recognition
16
  - vllm
17
  - multilingual
 
 
 
 
 
18
  ---
19
 
20
  <div align="center">
 
49
  </div>
50
  <br>
51
 
52
+ **Paper:** [Confucius4-R2T2: A Stable-Prefix Streaming ASR Model for Real Real-Time Transcription](https://huggingface.co/papers/2611.06043)
53
+
54
  Confucius4-R2T2 is a low-latency and high-accuracy true streaming Automatic Speech Recognition (ASR) model that features fine-grained and configurable decoding chunks from 80 ms to 2 s. The model operates in append-only output mode: committing transcript text permanently without revising previous words, which is critical for applications where text must be processed or acted upon instantly. This results in a smoother user experience, avoiding disruptive text revisions and visual flickering in real-time applications, such as Real-Time Live Captioning & Subtitling, Downstream NLP Pipelines & LLM Agents, Simultaneous Speech Translation, etc.
55
 
56
  R2T2, short for Real Real-Time Transcription, is built upon the Qwen3-ASR model. And it is trained with a unique set of data construction techniques including stable-prefix data, forced time-alignment data, and token-level audio segmentation. Combined with a Longest Stable Prefix (LSP) learning paradigm (tech report will be released soon), R2T2 can dynamically determine when a stable prefix can be safely emitted and when additional audio context is needed. By exposing only stable prefixes, the model provides high-quality context that conditions subsequent predictions while guaranteeing that previously emitted text remains unchanged. Despite its streaming design, R2T2 maintains strong accuracy in offline recognition.
 
727
  R2T2 uses **dual licensing** to distinguish the source code from the model weights:
728
 
729
  - **Code** in the accompanying GitHub repository is released under the [Apache License 2.0](https://github.com/netease-youdao/Confucius4-R2T2/blob/master/LICENSE) and is free to use, modify, and redistribute (including commercially) under the terms of that license.
730
+ - **Model weights** are released under the [NetEase Model Use License Agreement](https://github.com/netease-youdao/Confucius4-R2T2/blob/master/MODEL_LICENSE).