Elamur Research⌕
☰
home / papers / 2026 / Qwen3-ASR Technical Report

Qwen3-ASR Technical Report

Paper · 2026 · Xian Shi, Xiong Wang, Zhifang Guo et al.

Summary

The Qwen Team presents Qwen3-ASR, an open family of speech models built on the large audio-language model approach. In this approach, an LLM interprets audio before transcribing it, which helps with noise, long recordings, named entities and dialects. The family has two recognizers, Qwen3-ASR-1.7B and Qwen3-ASR-0.6B. Both perform language identification and transcription for 52 languages and dialects. A third model, Qwen3-ForcedAligner-0.6B, is a non-autoregressive, LLM-based tool that predicts word-, sentence- or paragraph-level timestamps.

The models are derived from Qwen3-Omni and pair an AuT audio encoder with a Qwen3 language model. Training runs in four stages: encoder pretraining, multimodal pretraining, supervised fine-tuning and GSPO reinforcement learning. Because public test sets are near their annotation-error ceiling, the authors also report internal benchmarks covering noise, dialects, children and elderly speakers. They find that the 1.7B model leads open-source ASR and rivals commercial APIs, while the 0.6B model offers the best accuracy-to-cost balance. The aligner outperforms MFA-style baselines such as NFA and WhisperX. All weights and the tooling are released under Apache 2.0.

Key points

  • Language coverage: 30 languages plus 22 Chinese dialects. Ukrainian is not supported, though Russian, Polish and Czech are. Accepted inputs include speech, singing and full songs with background music, at up to 20 minutes per file, in both streaming and offline modes.
  • Architecture:
    • The AuT encoder produces 12.5 Hz tokens and uses a dynamic attention window of 1–8 s.
    • The encoder has 300M parameters in the 1.7B model and 180M in the 0.6B model.
    • Encoder pretraining used about 40M hours of pseudo-labeled audio, multimodal pretraining used about 3T tokens, and the RL stage used about 50k utterances.
  • Efficiency (0.6B model): time-to-first-token as low as 92 ms. At a concurrency of 128 it transcribes roughly 2,000 s of audio per second (measured on vLLM).
  • Streaming quality (1.7B model): with 2 s chunks, average error rises only from 2.69 offline to 3.33 streaming on LibriSpeech and Fleurs.
  • Qwen3-ForcedAligner-0.6B:
    • Covers 11 languages and audio up to 300 s.
    • Average timestamp shift is about 43 ms on MFA-labeled data and 32 ms on human-labeled data, versus 100 ms or more for baselines.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksSignal Processing · 493 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,894 worksSpeech & spoken language · 519 works

Cite

Cite as Xian Shi, Xiong Wang, Zhifang Guo, Yongqi Wang, Pei Zhang, Xinyu Zhang, Zishan Guo, Hongkun Hao, Yu Xi, Baosong Yang, Jin Xu, Jingren Zhou and Junyang Lin. 2026. Qwen3-ASR Technical Report. arXiv:2601.21337.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2601.21337 · arXiv 2601.21337

@misc{shi2026qwen3,
  title = {{Qwen3-ASR Technical Report}},
  author = {Xian Shi, Xiong Wang, Zhifang Guo, Yongqi Wang, Pei Zhang, Xinyu Zhang, Zishan Guo, Hongkun Hao, Yu Xi, Baosong Yang, Jin Xu, Jingren Zhou and Junyang Lin},
  year = {2026},
  doi = {10.48550/arxiv.2601.21337},
  eprint = {2601.21337},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2601.21337}
}

References

23 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Keyu An, Yanni Chen, Zhigao Chen, Chong Deng, Zhihao Du, Changfeng Gao, Zhifu Gao, Bo Gong, Xiangang Li, Yabin Li, Ying Liu, Xiang Lv, Yunjie Ji, Yiheng Jiang, Bin Ma, Haoneng Luo, Chongjia Ni, Zexu Pan, Yiping Peng, Zhendong Peng, Peiyao Wang, Hao Wang, Haoxu Wang, Wen Wang, Wupeng Wang, Yuzhong Wu… — [arXiv:2509.12508](https://arxiv.org/abs/2509.12508) find
  2. Rosana Ardila, Megan Branson, Kelly Davis, Michael Kohler, Josh Meyer, Michael Henretty, Reuben Morais, Lindsay Saunders, Francis Tyers, and Gregor Weber. Common voice: A massively-multilingual speech corpus. In Proceedings of the twelfth language resources and evaluation conference, pp. 4218–4222,…
  3. Ye Bai, Jingping Chen, Jitong Chen, Wei Chen, Zhuo Chen, Chuang Ding, Linhao Dong, Qianqian Dong, Yujiao Du, Kepan Gao, Lu Gao, Yi Guo, Minglun Han, Ting Han, Wenchao Hu, Xinying Hu, Yuxiang Hu, Deyu Hua, Lu Huang, Mingkun Huang, Youjia Huang, Jishuo Jin, Fanliu Kong, Zongwei Lan, Tianyu Li, Xiaoyan… — [arXiv:2407.04675](https://arxiv.org/abs/2407.04675) find
  4. William Chan, Navdeep Jaitly, Quoc V. Le, and Oriol Vinyals. Listen, attend and spell: A neural network for large vocabulary conversational speech recognition. In Proc. ICASSP, pp. 4960–4964. IEEE, 2016. URL https://doi.org/10.1109/ICASSP.2016.7472621. ↗
  5. Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, et al. Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities, 2025. URL https:…
  6. Alexis Conneau, Min Ma, Simran Khanuja, Yu Zhang, Vera Axelrod, Siddharth Dalmia, Jason Riesa, Clara Rivera, and Ankur Bapna. Fleurs: Few-shot learning evaluation of universal representations of speech. In 2022 IEEE Spoken Language Technology Workshop (SLT), pp. 798–805.
  7. Yuhang Dai, Ziyu Zhang, Shuai Wang, Longhao Li, Zhao Guo, Tianlun Zuo, Shuiyuan Wang, Hongfei Xue, Chengyou Wang, Qing Wang, et al. Wenetspeech-chuan: A large-scale sichuanese corpus with rich annotation for dialectal speech processing. arXiv preprint arXiv:2509.18004, 2025. — [arXiv:2509.18004](https://arxiv.org/abs/2509.18004) find
  8. Alex Graves. Sequence Transduction with Recurrent Neural Networks, 2012. URL http://arxiv.org/ab s/1211.3711. ↗
  9. Ludwig Kürzinger, Dominik Winkelbauer, Lujun Li, Tobias Watzel, and Gerhard Rigoll. CTC-Segmentation of Large Corpora for German End-to-End Speech Recognition. In Proc. Speech and Computer, pp. 267–278, 2020. URL https://doi.org/10.1007/978-3-030-60276-5_27. — [doi:10.1007/978-3-030-60276-5_27](https://doi.org/10.1007/978-3-030-60276-5_27) ↗
  10. Longhao Li, Zhao Guo, Hongjie Chen, Yuhang Dai, Ziyu Zhang, Hongfei Xue, Tianlun Zuo, Chengyou Wang, Shuiyuan Wang, Jie Li, et al. Wenetspeech-yue: A large-scale cantonese speech corpus with multi-dimensional annotation. arXiv preprint arXiv:2509.03959, 2025. — [arXiv:2509.03959](https://arxiv.org/abs/2509.03959) find
  11. Michael McAuliffe, Michaela Socolof, Sarah Mihuc, Michael Wagner, and Morgan Sonderegger. Montreal Forced Aligner: Trainable Text-Speech Alignment Using Kaldi. In Proc. Interspeech, pp. 498–502, 2017. URL https://doi.org/10.21437/Interspeech.2017-1386. — [doi:10.21437/interspeech.2017-1386](https://doi.org/10.21437/interspeech.2017-1386) ↗
  12. Bingshen Mu, Pengcheng Guo, Zhaokai Sun, Shuai Wang, Hexin Liu, Mingchen Shao, Lei Xie, Eng Siong Chng, Longshuai Xiao, Qiangze Feng, and Daliang Wang. Summary on The Multilingual Conversational Speech Language Model Challenge: Datasets, Tasks, Baselines, and Methods. In Proc. ICASSP, 2026a. find
  13. Bingshen Mu, Xian Shi, Xiong Wang, Hexin Liu, Jin Xu, and Lei Xie. LLM-ForcedAligner: A Non-Autoregressive and Accurate LLM-Based Forced Aligner for Multilingual and Long-Form Speech, 2026b. URL https://arxiv.org/abs/2601.18220. ↗
  14. OpenAI. Hello GPT-4o, 2024. URL https://openai.com/index/hello-gpt-4o/. ↗
  15. Vassil Panayotov, Guoguo Chen, Daniel Povey, and Sanjeev Khudanpur. Librispeech: an asr corpus based on public domain audio books. In 2015 IEEE international conference on acoustics, speech and signal processing (ICASSP), pp. 5206–5210. find
  16. Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. Robust Speech Recognition via Large-Scale Weak Supervision. In Proc. ICML, pp. 28492–28518, 2023. URL https://proceedings.mlr.press/v202/radford23a.html. ↗
  17. Elena Rastorgueva, Vitaly Lavrukhin, and Boris Ginsburg. NeMo Forced Aligner and its application to word alignment for subtitle generation. In Proc. Interspeech, pp. 5257–5258, 2023. URL https://www.isca-archive.org/interspeech_2023/rastorgueva23_interspeech.html. ↗
  18. Xian Shi, Yanni Chen, Shiliang Zhang, and Zhijie Yan. Achieving timestamp prediction while recognizing with non-autoregressive end-to-end ASR model. In Proc. NCMMSC, 2023. find
  19. Jin Xu, Zhifang Guo, Hangrui Hu, Yunfei Chu, Xiong Wang, Jinzheng He, Yuxuan Wang, Xian Shi, Ting He, Xinfa Zhu, Yuanjun Lv, Yongqi Wang, Dake Guo, He Wang, Linhan Ma, Pei Zhang, Xinyu Zhang, Hongkun Hao, Zishan Guo, Baosong Yang, Bin Zhang, Ziyang Ma, Xipin Wei, Shuai Bai, Keqin Chen, Xuejing Liu,… — Qwen3-Omni Technical Report
  20. Jin Xu, Zhifang Guo, Hangrui Hu, Yunfei Chu, Xiong Wang, Jinzheng He, Yuxuan Wang, Xian Shi, Ting He, Xinfa Zhu, Yuanjun Lv, Yongqi Wang, Dake Guo, He Wang, Linhan Ma, Pei Zhang, Xinyu Zhang, Hongkun Hao, Zishan Guo, Baosong Yang, Bin Zhang, Ziyang Ma, Xipin Wei, Shuai Bai, Keqin Chen, Xuejing Liu,… — Qwen3-Omni Technical Report
  21. Z.ai. Glm asr 2512. https://docs.z.ai/guides/audio/glm-asr-2512, 2025. Accessed: 2026-01-26. ↗
  22. Binbin Zhang, Hang Lv, Pengcheng Guo, Qijie Shao, Chao Yang, Lei Xie, Xin Xu, Hui Bu, Xiaoyu Chen, Chenchen Zeng, et al. Wenetspeech: A 10000+ hours multi-domain mandarin corpus for speech recognition. In Proc. ICASSP, pp. 6182–6186, 2022. find
  23. Chujie Zheng, Shixuan Liu, Mingze Li, Xiong-Hui Chen, Bowen Yu, Chang Gao, Kai Dang, Yuqiong Liu, Rui Men, An Yang, Jingren Zhou, and Junyang Lin. Group Sequence Policy Optimization, 2025. URL https://arxiv.org/abs/2507.18071. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/shi-2026-qwen3-asr-technical-report/