---
title: "Vietnamese Automatic Speech Recognition: A Revisit (2026)"
description: "The authors (Qualcomm AI Research) design a seven-step, language-agnostic pipeline that merges such sources: audio is… 26 references."
authors: ["Thi Vu", "Linh The Nguyen", "Dat Quoc Nguyen"]
year: 2026
venue: "Findings of the Association for Computational Linguistics: EACL 2026"
source: "https://aclanthology.org/2026.findings-eacl.345/"
doi: "10.18653/v1/2026.findings-eacl.345"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/vu-2026-vietnamese-automatic-speech-recognition-revisit/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/vu-2026-vietnamese-automatic-speech-recognition-revisit/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/47/) / Vietnamese Automatic Speech Recognition: A Revisit

# Vietnamese Automatic Speech Recognition: A Revisit

**Authors:** Thi Vu, [Linh The Nguyen](https://md.research.elamur.ai/free/en/authors/linh-the-nguyen/) and [Dat Quoc Nguyen](https://md.research.elamur.ai/free/en/authors/dat-quoc-nguyen/) · **Year:** 2026 · **Venue:** [Findings of the Association for Computational Linguistics: EACL 2026](https://md.research.elamur.ai/free/en/venues/findings/) ([Findings of the Association for Computational Linguistics 2026](https://md.research.elamur.ai/free/en/venues/findings/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.findings-eacl.345/) · [PDF](https://aclanthology.org/2026.findings-eacl.345.pdf) · [DOI 10.18653/v1/2026.findings-eacl.345](https://doi.org/10.18653/v1/2026.findings-eacl.345) · [OpenAlex](https://openalex.org/W7140116537)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
Open Vietnamese speech corpora are fragmented. Read-speech sets such as VIVOS and CMV-vi-14 are clean but small and narrow. Spontaneous collections like BUD500 or ViMD have uneven transcript quality and mixed formats and sampling rates. Almost none provide punctuation, casing or word-level timestamps.

The authors (Qualcomm AI Research) design a seven-step, language-agnostic pipeline that merges such sources:
- audio is cleaned and resampled;
- transcripts are generated or validated by comparing independent ASR outputs and dropping pairs above a WER threshold;
- a neural model restores punctuation and capitalization;
- numbers are temporarily spelled out so that a CTC-based forced aligner in the WhisperX style can assign word timings, after which digits are restored.

Applied to Vietnamese, the pipeline yields a unified 500-hour corpus. A whisper-small model fine-tuned on 469 hours of it outperforms PhoWhisper-small, which used 844 hours, especially on orthographic WER, and it also predicts timestamps. The work suggests that careful curation can matter more than raw data volume. It offers a reusable recipe for other low-resource languages, though the filtering may bias the data toward easier acoustic conditions.

## Key points
- Pipeline steps:
  1. collect and clean the audio;
  2. create transcripts via agreement between two ASR models;
  3. filter existing transcripts against a secondary ASR model;
  4. restore punctuation and casing;
  5. expand numbers to words;
  6. align timestamps;
  7. convert numbers back to digits.
- PhoASR-whisper-small-469h reaches O-WER 12.46% and N-WER 8.69%. For comparison:
  - pre-trained whisper-small: 70.16% / 64.07%;
  - wav2vec2 (469h): 51.15% / 14.10%;
  - PhoWhisper-small: 33.90% / 8.97%.
- PhoASR-whisper-small-3100h mixes the curated data with larger, minimally processed corpora and was trained for 15 epochs. It scores O-WER 11.70% and N-WER 8.20%. Timestamp F1 rises from 76.40% to 83.68%, and IoU from 55.62 to 57.39.
- ChunkFormer-large-vi, trained on 25K private hours, has the best N-WER (6.89%). Its O-WER of 32.43%, however, indicates weak punctuation and capitalization.
- The resources are available via the PhoASR project page (github.com/qualcomm-ai-research/PhoASR). The paper appeared in Findings of EACL 2026.

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

## Institutions

- [Qualcomm](https://md.research.elamur.ai/free/en/companies/qualcomm/) — Company, San Diego, United States · 6 works · 144 citations

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [🔊 Signal Processing](https://md.research.elamur.ai/free/en/shelves/computer-science/signal-processing/) · 452 works

🧭 Topics (1): [🎙️ Speech & spoken language](https://md.research.elamur.ai/free/en/topics/speech/) · 480 works

## Cite

Cite as: Thi Vu, Linh The Nguyen and Dat Quoc Nguyen. 2026. Vietnamese Automatic Speech Recognition: A Revisit. Findings of the Association for Computational Linguistics: EACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{vu2026vietnamese,
  title = {{Vietnamese Automatic Speech Recognition: A Revisit}},
  author = {Thi Vu, Linh The Nguyen and Dat Quoc Nguyen},
  booktitle = {Findings of the Association for Computational Linguistics: EACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-eacl.345},
  url = {https://doi.org/10.18653/v1/2026.findings-eacl.345}
}
```

- [← previous VIDA: A Visual Intent-driven Design Assistant for Proactive Multimodal Clarification](https://md.research.elamur.ai/free/en/papers/liu-2026-vida-visual-intent-driven-design/)
- [next → VIGiA: Instructional Video Guidance via Dialogue Reasoning and Retrieval](https://md.research.elamur.ai/free/en/papers/silva-2026-vigia-instructional-video-guidance-dialogue/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/47/)

## References

26 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Rosana Ardila, Megan Branson, Kelly Davis, Michael Kohler, Josh Meyer, Michael Henretty, Reuben Morais, Lindsay Saunders, Francis Tyers, and Gregor Weber. 2020. [Common Voice: A Massively-Multilingual Speech Corpus](https://md.research.elamur.ai/free/en/papers/ardila-2019-common-voice-massively-multilingual-speech/). In Proceedings of LREC, pages 4211–4215.
2. Alexei Baevski, Henry Zhou, Abdelrahman Mohamed, and Michael Auli. 2020. [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://md.research.elamur.ai/free/en/papers/baevski-2020-wav2vec-2-0-framework-self/). In Proceedings of NeurIPS.
3. Max Bain, Jaesung Huh, Tengda Han, and Andrew Zisserman. 2023. WhisperX: Time-Accurate Speech Transcription of Long-Form Audio. Proceedings of INTERSPEECH, pages 4489–4493. [find](https://scholar.google.com/scholar?q=Max+Bain%2C+Jaesung+Huh%2C+Tengda+Han%2C+and+Andrew+Zisserman%2E+2023%2E+WhisperX%3A+Time%2DAccurate+Speech+Transcription+of+Long%2DForm+Audio%2E+Proceedings+of+INTERSPEECH%2C+pages+4489%E2%80%934493%2E)
4. Guoguo Chen, Shuzhou Chai, Guan-Bo Wang, Jiayu Du, Wei-Qiang Zhang, Chao Weng, Dan Su, Daniel Povey, Jan Trmal, Junbo Zhang, Mingjie Jin, Sanjeev Khudanpur, Shinji Watanabe, Shuaijiang Zhao, Wei Zou, Xiangang Li, Xuchen Yao, Yongqing Wang, Zhao You, and Zhiyong Yan. 2021. GigaSpeech: An Evolving, Mu… [find](https://scholar.google.com/scholar?q=Guoguo+Chen%2C+Shuzhou+Chai%2C+Guan%2DBo+Wang%2C+Jiayu+Du%2C+Wei%2DQiang+Zhang%2C+Chao+Weng%2C+Dan+Su%2C+Daniel+Povey%2C+Jan+Trmal%2C+Junbo+Zhang%2C+Mingjie+Jin%2C+Sanjeev+Khudanpur%2C+Shinji+Watanabe%2C+Shuaijiang+Zhao%2C+Wei+Zou%2C+)
5. Nguyen Van Dinh, Thanh Chi Dang, Luan Thanh Nguyen, and Kiet Van Nguyen. 2024. Multi-Dialect Vietnamese: Task, Dataset, Baseline Models and Challenges. In Proceedings of EMNLP, pages 7476–7498. [find](https://scholar.google.com/scholar?q=Nguyen+Van+Dinh%2C+Thanh+Chi+Dang%2C+Luan+Thanh+Nguyen%2C+and+Kiet+Van+Nguyen%2E+2024%2E+Multi%2DDialect+Vietnamese%3A+Task%2C+Dataset%2C+Baseline+Models+and+Challenges%2E+In+Proceedings+of+EMNLP%2C+pages+7476%E2%80%937498%2E)
6. Le Duy Khanh. 2022. Finetune Wav2vec 2.0 For Vietnamese Speech Recognition. [find](https://scholar.google.com/scholar?q=Le+Duy+Khanh%2E+2022%2E+Finetune+Wav2vec+2%2E0+For+Vietnamese+Speech+Recognition%2E)
7. Daniel Galvez, Greg Diamos, Juan Manuel Ciro Torres, Juan Felipe Cerón, Keith Achorn, Anjali Gopi, David Kanter, Max Lam, Mark Mazumder, and Vijay Janapa Reddi. 2021. The People’s Speech: A Large-Scale Diverse English Speech Recognition Dataset for Commercial Usage. In Proceedings of NeurIPS Dataset… [find](https://scholar.google.com/scholar?q=Daniel+Galvez%2C+Greg+Diamos%2C+Juan+Manuel+Ciro+Torres%2C+Juan+Felipe+Cer%C3%B3n%2C+Keith+Achorn%2C+Anjali+Gopi%2C+David+Kanter%2C+Max+Lam%2C+Mark+Mazumder%2C+and+Vijay+Janapa+Reddi%2E+2021%2E+The+People%E2%80%99s+Speech%3A+A+Large%2DScal)
8. Thinh Le Phuoc Gia, Tuan Pham Minh, Hung Nguyen Quoc, Trung Nguyen Quoc, and Vinh Truong Hoang. 2024. viVoice: Enabling Vietnamese Multi-Speaker Speech Synthesis. [find](https://scholar.google.com/scholar?q=Thinh+Le+Phuoc+Gia%2C+Tuan+Pham+Minh%2C+Hung+Nguyen+Quoc%2C+Trung+Nguyen+Quoc%2C+and+Vinh+Truong+Hoang%2E+2024%2E+viVoice%3A+Enabling+Vietnamese+Multi%2DSpeaker+Speech+Synthesis%2E)
9. Hugging Face. 2024. Unit 5: Evaluation metrics for asr. [https://huggingface.co/learn/](https://huggingface.co/learn/) Ac-audio-course/en/chapter5/evaluation. cessed: 2024-07-29. [↗](https://huggingface.co/learn/)
10. Khanh Le, Tuan Vu Ho, Dung Tran, and Duc Thanh Chau. 2025. ChunkFormer: Masked Chunking Conformer For Long-Form Speech Transcription. In Proceedings of ICASSP. [find](https://scholar.google.com/scholar?q=Khanh+Le%2C+Tuan+Vu+Ho%2C+Dung+Tran%2C+and+Duc+Thanh+Chau%2E+2025%2E+ChunkFormer%3A+Masked+Chunking+Conformer+For+Long%2DForm+Speech+Transcription%2E+In+Proceedings+of+ICASSP%2E)
11. Thanh-Thien Le, Linh The Nguyen, and Dat Quoc Nguyen. 2024. PhoWhisper: Automatic Speech Recognition for Vietnamese. In Proceedings of the ICLR 2024 Tiny Papers track. [find](https://scholar.google.com/scholar?q=Thanh%2DThien+Le%2C+Linh+The+Nguyen%2C+and+Dat+Quoc+Nguyen%2E+2024%2E+PhoWhisper%3A+Automatic+Speech+Recognition+for+Vietnamese%2E+In+Proceedings+of+the+ICLR+2024+Tiny+Papers+track%2E)
12. Khai Le-Duc. 2024. VietMed: A Dataset and Benchmark for Automatic Speech Recognition of Vietnamese in the Medical Domain. In Proceedings of LREC-COLING, pages 17365–17370. [find](https://scholar.google.com/scholar?q=Khai+Le%2DDuc%2E+2024%2E+VietMed%3A+A+Dataset+and+Benchmark+for+Automatic+Speech+Recognition+of+Vietnamese+in+the+Medical+Domain%2E+In+Proceedings+of+LREC%2DCOLING%2C+pages+17365%E2%80%9317370%2E)
13. Hieu-Thi Luong and Hai-Quan Vu. 2016. A non-expert Kaldi recipe for Vietnamese Speech Recognition System. In Proceedings of WLSI/OIAF4HLT, pages 51– 55. [find](https://scholar.google.com/scholar?q=Hieu%2DThi+Luong+and+Hai%2DQuan+Vu%2E+2016%2E+A+non%2Dexpert+Kaldi+recipe+for+Vietnamese+Speech+Recognition+System%2E+In+Proceedings+of+WLSI%2FOIAF4HLT%2C+pages+51%E2%80%93+55%2E)
14. Michael McAuliffe, Michaela Socolof, Sarah Mihuc, Michael Wagner, and Morgan Sonderegger. 2017. Montreal Forced Aligner: Trainable Text-Speech Alignment Using Kaldi. In Proceedings of INTER-SPEECH, pages 498–502. [find](https://scholar.google.com/scholar?q=Michael+McAuliffe%2C+Michaela+Socolof%2C+Sarah+Mihuc%2C+Michael+Wagner%2C+and+Morgan+Sonderegger%2E+2017%2E+Montreal+Forced+Aligner%3A+Trainable+Text%2DSpeech+Alignment+Using+Kaldi%2E+In+Proceedings+of+INTER%2DSPEECH%2C+pa)
15. Vu Le Binh Nguyen. 2021. Wav2Vec2 Base Vietnamese Model. [https://huggingface.co/](https://huggingface.co/) nguyenvulebinh/wav2vec2-base-vi. [↗](https://huggingface.co/)
16. Pham Quang Nhut, Duong Pham Hoang Anh, and Nguyen Vinh Tiep. 2024. VSV-1100: Vietnamese social voice dataset. [find](https://scholar.google.com/scholar?q=Pham+Quang+Nhut%2C+Duong+Pham+Hoang+Anh%2C+and+Nguyen+Vinh+Tiep%2E+2024%2E+VSV%2D1100%3A+Vietnamese+social+voice+dataset%2E)
17. Vassil Panayotov, Guoguo Chen, Daniel Povey, and Sanjeev Khudanpur. 2015. Librispeech: An ASR corpus based on public domain audio books. In Proceedings of ICASSP, pages 5206–5210. [find](https://scholar.google.com/scholar?q=Vassil+Panayotov%2C+Guoguo+Chen%2C+Daniel+Povey%2C+and+Sanjeev+Khudanpur%2E+2015%2E+Librispeech%3A+An+ASR+corpus+based+on+public+domain+audio+books%2E+In+Proceedings+of+ICASSP%2C+pages+5206%E2%80%935210%2E)
18. Douglas B. Paul and Janet M. Baker. 1992. The design for the Wall Street Journal-based CSR corpus. In Proceedings of HLT, pages 357–362. [find](https://scholar.google.com/scholar?q=Douglas+B%2E+Paul+and+Janet+M%2E+Baker%2E+1992%2E+The+design+for+the+Wall+Street+Journal%2Dbased+CSR+corpus%2E+In+Proceedings+of+HLT%2C+pages+357%E2%80%93362%2E)
19. Anh Pham, Khanh Linh Tran, Linh Nguyen, Thanh Duy Cao, Phuc Phan, and Duong A. Nguyen. 2024. Bud500: A Comprehensive Vietnamese ASR Dataset. [find](https://scholar.google.com/scholar?q=Anh+Pham%2C+Khanh+Linh+Tran%2C+Linh+Nguyen%2C+Thanh+Duy+Cao%2C+Phuc+Phan%2C+and+Duong+A%2E+Nguyen%2E+2024%2E+Bud500%3A+A+Comprehensive+Vietnamese+ASR+Dataset%2E)
20. Viet Thanh Pham, Xuan Thai Hoa Nguyen, Vu Hoang, and Thi Thu Trang Nguyen. 2023. Vietnam-Celeb: A large-scale dataset for Vietnamese speaker recognition. In Proceedings of INTERSPEECH, pages 1918–1922. [find](https://scholar.google.com/scholar?q=Viet+Thanh+Pham%2C+Xuan+Thai+Hoa+Nguyen%2C+Vu+Hoang%2C+and+Thi+Thu+Trang+Nguyen%2E+2023%2E+Vietnam%2DCeleb%3A+A+large%2Dscale+dataset+for+Vietnamese+speaker+recognition%2E+In+Proceedings+of+INTERSPEECH%2C+pages+1918%E2%80%931922)
21. Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2023. [Robust speech recognition via large-scale weak supervision](https://md.research.elamur.ai/free/en/papers/radford-2022-robust-speech-recognition-large-scale/). In Proceedings of ICML.
22. Duc Chung Tran. 2020. FPT Open Speech Dataset (FOSD) - Vietnamese. [find](https://scholar.google.com/scholar?q=Duc+Chung+Tran%2E+2020%2E+FPT+Open+Speech+Dataset+%28FOSD%29+%2D+Vietnamese%2E)
23. Linh Thi Thuc Tran, Han-Gyu Kim, Hoang Minh La, and Su Van Pham. 2024. Automatic Speech Recognition of Vietnamese for a New Large-Scale Corpus. Electronics, 13(5):977. [find](https://scholar.google.com/scholar?q=Linh+Thi+Thuc+Tran%2C+Han%2DGyu+Kim%2C+Hoang+Minh+La%2C+and+Su+Van+Pham%2E+2024%2E+Automatic+Speech+Recognition+of+Vietnamese+for+a+New+Large%2DScale+Corpus%2E+Electronics%2C+13%285%29%3A977%2E)
24. Nguyen Luong Tran, Duong Minh Le, and Dat Quoc Nguyen. 2022. BARTpho: Pre-trained Sequence-to-Sequence Models for Vietnamese. In Proceedings of INTERSPEECH. [find](https://scholar.google.com/scholar?q=Nguyen+Luong+Tran%2C+Duong+Minh+Le%2C+and+Dat+Quoc+Nguyen%2E+2022%2E+BARTpho%3A+Pre%2Dtrained+Sequence%2Dto%2DSequence+Models+for+Vietnamese%2E+In+Proceedings+of+INTERSPEECH%2E)
25. Christophe Veaux, Junichi Yamagishi, and Kirsten Mac-Donald. 2017. CSTR VCTK corpus: English multi-speaker corpus for CSTR voice cloning toolkit (version 0.92). [find](https://scholar.google.com/scholar?q=Christophe+Veaux%2C+Junichi+Yamagishi%2C+and+Kirsten+Mac%2DDonald%2E+2017%2E+CSTR+VCTK+corpus%3A+English+multi%2Dspeaker+corpus+for+CSTR+voice+cloning+toolkit+%28version+0%2E92%29%2E)
26. Thi Vu, Linh The Nguyen, and Dat Quoc Nguyen. 2025. Zero-Shot Text-to-Speech for Vietnamese. In Proceedings of ACL. [find](https://scholar.google.com/scholar?q=Thi+Vu%2C+Linh+The+Nguyen%2C+and+Dat+Quoc+Nguyen%2E+2025%2E+Zero%2DShot+Text%2Dto%2DSpeech+for+Vietnamese%2E+In+Proceedings+of+ACL%2E)

*Updated 1 November 2026*