---
title: "AfriSpeech-MultiBench — paper by Ashungafac et al. (2025)"
description: "The authors build AfriSpeech-MultiBench, an evaluation suite that combines seven African-accented English… 5% WER. 31 references."
authors: ["Gabrial Zencha Ashungafac", "Mardhiyah Sanni", "Busayo Awobade", "Alex Gichamba", "Tobi Olatunji"]
year: 2025
venue: "Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics"
source: "https://aclanthology.org/2025.ijcnlp-long.190/"
doi: "10.18653/v1/2025.ijcnlp-long.190"
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/ashungafac-2025-afrispeech-multibench-verticalized-multidomain-multicountry/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/ashungafac-2025-afrispeech-multibench-verticalized-multidomain-multicountry/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2025](https://md.research.elamur.ai/free/en/papers/2025/page/16/) / AfriSpeech-MultiBench: A Verticalized Multidomain Multicountry Benchmark Suite for African Accented English ASR

# AfriSpeech-MultiBench: A Verticalized Multidomain Multicountry Benchmark Suite for African Accented English ASR

**Authors:** Gabrial Zencha Ashungafac, [Mardhiyah Sanni](https://md.research.elamur.ai/free/en/authors/mardhiyah-sanni/), Busayo Awobade, Alex Gichamba and [Tobi Olatunji](https://md.research.elamur.ai/free/en/authors/tobi-olatunji/) · **Year:** 2025 · **Venue:** [Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics](https://md.research.elamur.ai/free/en/venues/ijcnlp/) ([IJCNLP 2025](https://md.research.elamur.ai/free/en/venues/ijcnlp/2025/)) · **Publisher:** [The Asian Federation of Natural Language Processing and The Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/the-asian-federation-of-natural-language-processing-and-the-association-for/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2025.ijcnlp-long.190/) · [PDF](https://aclanthology.org/2025.ijcnlp-long.190.pdf) · [DOI 10.18653/v1/2025.ijcnlp-long.190](https://doi.org/10.18653/v1/2025.ijcnlp-long.190) · [OpenAlex](https://openalex.org/W7138169528)

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

## Summary
Leaderboards for English speech recognition rely mostly on North-American and British speech, so they say little about how systems handle African-accented English in real applications. The authors build AfriSpeech-MultiBench, an evaluation suite that combines seven African-accented English corpora (read, conversational, parliamentary, clinical, entity-rich and call-center audio). The data are grouped into seven "vertical" domains: a robustness/hallucination diagnostic, plus Finance, Legal, Call Center, Medical, Named Entities and General speech. Eighteen recognizers are tested zero-shot and scored by WER. They span Conformer/transducer models, Whisper variants, speech-augmented LLMs and commercial cloud APIs. Results are also broken down by country and accent. Models that score below 5% WER on standard benchmarks can do 5–10 times worse here, especially on conversations, names, numbers and financial commands. The regionally trained Intron-Sahara models rank best almost everywhere. Hallucination on silence and missing speech remains common. The benchmark is public on Hugging Face (CC BY-NC-SA 4.0). It lets practitioners choose systems for African deployments based on evidence from the right domain, not on global averages.

## Key points
- The suite merges AfriSpeech-200, AfriSpeech-Dialog, AfriSpeech-Parliamentary, Med-Conv-Nig, AfriNames, AfriSpeech-Countries and a private Afro-Call-Centers set. In total it covers about 79 hours, 100+ accents and 10+ countries.
- The robustness domain has three subsets:
  - short clips under 3.5 s, drawn from AMI, VoxPopuli, AfriSpeech and AfriNames;
  - AfriNames clips with deliberately inserted pauses;
  - a no-speech subset from AfriSpeech-Parliament that measures false triggering.
- Average WER across countries in the per-accent table:
  - Intron-Sahara V2: about 13.9%;
  - Intron-Sahara: about 15.1%;
  - Whisper-large-v3: about 21.2%;
  - GPT-4o Transcribe: about 38.9%;
  - CrisperWhisper: about 66.2%.
- Most models exceed 40% WER on AfriNames names, numbers and voice commands.
- Neither model size nor architecture predicted reliability. Open models tended to get worse as they grew. Smaller global leaders such as Parakeet-tdt-0.6B and Distil-Whisper gave uneven results on African data.

## Interesting ideas
- Instead of one pooled test set, the evaluation is sliced by application vertical, so a model is judged on the domain it will actually serve; averages hide large domain-specific failures.
- A diagnostic subset contains audio with no speech or long pauses, where the correct output is empty or unchanged, directly measuring hallucination.
- Entity- and number-dense subsets (currencies, decimals, dates, transaction commands) are separated out because they break systems that do well on ordinary text.
- Results are disaggregated by country and accent cluster, revealing that a model's ranking can flip across regions.
- The authors openly flag that some sets are only proxies for their vertical, such as parliament debates standing in for legal speech.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (learn): Report parser and reader scores separately per genre (fables, school tests, maths problems, news) so a regression in one vertical cannot be masked by gains elsewhere.
- **I2** (learn): Add negative-control items whose correct answer is "nothing / not stated" (e.g. questions not answerable from the story) to catch teacher-induced or rule-induced fabrication.
- **I3** (tokens): Build a dedicated tokenizer/lemma stress set of money amounts, decimals, dates and names to test segmentation and tagging in isolation.
- **I4** (tree): Score LAS separately per English variety and per UD treebank source, so dialect conversion problems show up as localized drops rather than a blurred average.
- **I5** (answer): Label each evaluation set with how faithfully it represents the target task, so conclusions about reader ability are not overstated from proxy data.

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

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works · [🔊 Signal Processing](https://md.research.elamur.ai/free/en/shelves/computer-science/signal-processing/) · 493 works

🧭 Topics (3): [English](https://md.research.elamur.ai/free/en/topics/english/) · 2,379 works · [📏 Evaluation & benchmarks](https://md.research.elamur.ai/free/en/topics/evaluation-benchmarks/) · 1,341 works · [🎙️ Speech & spoken language](https://md.research.elamur.ai/free/en/topics/speech/) · 519 works

## Cite

Cite as: Gabrial Zencha Ashungafac, Mardhiyah Sanni, Busayo Awobade, Alex Gichamba and Tobi Olatunji. 2025. AfriSpeech-MultiBench: A Verticalized Multidomain Multicountry Benchmark Suite for African Accented English ASR. Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics.

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

```bibtex
@inproceedings{ashungafac2025afrispeech,
  title = {{AfriSpeech-MultiBench: A Verticalized Multidomain Multicountry Benchmark Suite for African Accented English ASR}},
  author = {Gabrial Zencha Ashungafac, Mardhiyah Sanni, Busayo Awobade, Alex Gichamba and Tobi Olatunji},
  booktitle = {Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics},
  year = {2025},
  doi = {10.18653/v1/2025.ijcnlp-long.190},
  url = {https://doi.org/10.18653/v1/2025.ijcnlp-long.190}
}
```

- [← previous Accusation, Anger, and Defense: Rhetorical Questions in Genesis, Exodus, Numbers, and Judges](https://md.research.elamur.ai/free/en/papers/nabulsi-2025-accusation-anger-defense-rhetorical-questions/)
- [next → Aged to Perfection: Machine-Learning Maps of Age in Conversational English](https://md.research.elamur.ai/free/en/papers/tang-2025-aged-perfection-machine-learning-maps/)
- [↑ back to papers of 2025](https://md.research.elamur.ai/free/en/papers/2025/page/16/)

## References

31 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Marah Abdin, Jyoti Aneja, Harkirat Behl, Sébastien Bubeck, Ronen Eldan, Suriya Gunasekar, Michael Harrison, Russell J. Hewett, Mojan Javaheripi, Piero Kauffmann, James R. Lee, Yin Tat Lee, Yuanzhi Li, Weishung Liu, Caio C. T. Mendes, Anh Nguyen, Eric Price, Gustavo de Rosa, Olli Saarikivi, and 8 others. 2024. [Phi-4 technical report](https://md.research.elamur.ai/free/en/papers/abdin-2024-phi-4-technical-report/). Preprint, arXiv:2412.08905. [↗](https://arxiv.org/abs/2412.08905)
2. David Ifeoluwa Adelani, Jessica Ojo, Israel Abebe Azime, Jian Yun Zhuang, Jesujoba Oluwadara Alabi, Xuanli He, and 1 others. 2025. Irokobench: A benchmark for african languages in the age of large language models. In Proceedings of the 2025 Conference of the North American Chapter of the Association for Computational Linguistics (NAACL), pages 2732–2757, Mexico City, Mexico. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=David+Ifeoluwa+Adelani%2C+Jessica+Ojo%2C+Israel+Abebe+Azime%2C+Jian+Yun+Zhuang%2C+Jesujoba+Oluwadara+Alabi%2C+Xuanli+He%2C+and+1+others%2E+2025%2E+Irokobench%3A+A+benchmark+for+african+languages+in+the+age+of+large+lan)
3. Galen Andrew, Mingqing Chen, Jinyu Lu, and Kevin Sim. 2022. Earnings22: A 100-hour benchmark corpus for earnings-call ASR. In Proceedings of Interspeech 2022, pages 3158–3162. [find](https://scholar.google.com/scholar?q=Galen+Andrew%2C+Mingqing+Chen%2C+Jinyu+Lu%2C+and+Kevin+Sim%2E+2022%2E+Earnings22%3A+A+100%2Dhour+benchmark+corpus+for+earnings%2Dcall+ASR%2E+In+Proceedings+of+Interspeech+2022%2C+pages+3158%E2%80%933162%2E)
4. Rosana Ardila, Megan Branson, Kelly Davis, Michael Kohler, Josh Meyer, Michael Henretty, and 1 others. 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 the 12th Language Resources and Evaluation Conference (LREC), pages 4218–4222.
5. 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 Advances in Neural Information Processing Systems (NeurIPS), pages 12449–12460.
6. Jean Carletta, Simone Ashby, Séverine Bourban, Mike Flynn, Mael Guillemot, Thomas Hain, and 1 others. 2005. The AMI meeting corpus: A pre-announcement. In Proceedings of the Second International Conference on Machine Learning for Multimodal Interaction (MLMI), pages 30–44. Springer. [find](https://scholar.google.com/scholar?q=Jean+Carletta%2C+Simone+Ashby%2C+S%C3%A9verine+Bourban%2C+Mike+Flynn%2C+Mael+Guillemot%2C+Thomas+Hain%2C+and+1+others%2E+2005%2E+The+AMI+meeting+corpus%3A+A+pre%2Dannouncement%2E+In+Proceedings+of+the+Second+International+Confe)
7. Chang Chen, Yiming Peng, Yuan Guo, Nanxin Yang, Shuai Zhang, Yongqiang Cui, and 1 others. 2021. Gigaspeech: An evolving, multi-domain ASR training corpus with 10,000 hours of audio. In Proceedings of Interspeech 2021, pages 3790–3794. [find](https://scholar.google.com/scholar?q=Chang+Chen%2C+Yiming+Peng%2C+Yuan+Guo%2C+Nanxin+Yang%2C+Shuai+Zhang%2C+Yongqiang+Cui%2C+and+1+others%2E+2021%2E+Gigaspeech%3A+An+evolving%2C+multi%2Ddomain+ASR+training+corpus+with+10%2C000+hours+of+audio%2E+In+Proceedings+of+)
8. Zhehuai Chen, He Huang, Andrei Andrusenko, Oleksii Hrinchuk, Krishna Puvvada, Jason Li, and 1 others. 2023. SALM: Speech-augmented language model with in-context learning for speech recognition and translation. arXiv preprint arXiv:2310.09424. [↗](https://arxiv.org/abs/2310.09424)
9. Alex DiChristofano, Henry Shuster, Shefali Chandra, and Neal Patwari. 2022. [Global performance disparities between english-language accents in automatic speech recognition](https://md.research.elamur.ai/free/en/papers/dichristofano-2022-global-performance-disparities-between-english/). In arXiv preprint arXiv:2208.01157. [↗](https://arxiv.org/abs/2208.01157)
10. Daniel Galvez, Vladimir Bataev, Hainan Xu, and Tim Kaldewey. 2024. Speed of light exact greedy decoding for rnn-t speech recognition models on gpu. In Interspeech 2024, pages 277–281. [find](https://scholar.google.com/scholar?q=Daniel+Galvez%2C+Vladimir+Bataev%2C+Hainan+Xu%2C+and+Tim+Kaldewey%2E+2024%2E+Speed+of+light+exact+greedy+decoding+for+rnn%2Dt+speech+recognition+models+on+gpu%2E+In+Interspeech+2024%2C+pages+277%E2%80%93281%2E)
11. Cong Guo, Jing Zhang, Xiaohui Ma, Yongqiang Huang, Mike Lewis, Zhe Wei, and Shiliang Chen. 2022. SPGISpeech: 5,000 hours of transcribed financial audio for self-supervised speech representation learning. In Proceedings of Interspeech 2022, pages 3663– 3667. [find](https://scholar.google.com/scholar?q=Cong+Guo%2C+Jing+Zhang%2C+Xiaohui+Ma%2C+Yongqiang+Huang%2C+Mike+Lewis%2C+Zhe+Wei%2C+and+Shiliang+Chen%2E+2022%2E+SPGISpeech%3A+5%2C000+hours+of+transcribed+financial+audio+for+self%2Dsupervised+speech+representation+learni)
12. François Hernandez, Vincent Nguyen, Sahar Ghannay, Natalia Tomashenko, and Yannick Estève. 2018. TED-LIUM 3: Twice as much data and corpus repartition for experiments on speaker adaptation. In Speech and Computer (SPECOM), pages 198–208. [find](https://scholar.google.com/scholar?q=Fran%C3%A7ois+Hernandez%2C+Vincent+Nguyen%2C+Sahar+Ghannay%2C+Natalia+Tomashenko%2C+and+Yannick+Est%C3%A8ve%2E+2018%2E+TED%2DLIUM+3%3A+Twice+as+much+data+and+corpus+repartition+for+experiments+on+speaker+adaptation%2E+In+Speech+)
13. Intron Health. 2025. Afrispeech-countries: Cross-regional african-accented english speech benchmark. [https://huggingface.co/datasets/intronhealth/](https://huggingface.co/datasets/intronhealth/) afrispeech-countries. [↗](https://huggingface.co/datasets/intronhealth/)
14. Intron Health. 2025a. Afrispeech-parliament: Transcribed parliamentary sessions from four african nations. [https://huggingface.co/datasets/](https://huggingface.co/datasets/) intronhealth/afrispeech-parliament. [↗](https://huggingface.co/datasets/)
15. Intron Health. 2025b. Afri-names: African named-entity read-speech corpus. [https://huggingface.co/datasets/intronhealth/](https://huggingface.co/datasets/intronhealth/) afri-names. [↗](https://huggingface.co/datasets/intronhealth/)
16. Intron Health. 2025c. Med-convo-nig: Nigerian doctor–patient tele-consultation speech dataset. [https://huggingface.co/datasets/](https://huggingface.co/datasets/) intronhealth/med-convo-nig. [↗](https://huggingface.co/datasets/)
17. Allison Koenecke, Andrew Nam, Emily Lake, Joe Nudell, Minnie Quartey, Zion Mengesha, and 1 others. 2020. Racial disparities in automated speech recognition. Proceedings of the National Academy of Sciences, 117(14):7684–7689. [find](https://scholar.google.com/scholar?q=Allison+Koenecke%2C+Andrew+Nam%2C+Emily+Lake%2C+Joe+Nudell%2C+Minnie+Quartey%2C+Zion+Mengesha%2C+and+1+others%2E+2020%2E+Racial+disparities+in+automated+speech+recognition%2E+Proceedings+of+the+National+Academy+of+Scie)
18. Alexander H. Liu, Andy Ehrenberg, Andy Lo, Clément Denoix, Corentin Barreau, Guillaume Lample, Jean-Malo Delignon, Khyathi Raghavi Chandu, Patrick von Platen, Pavankumar Reddy Muddireddy, Sanchit Gandhi, Soham Ghosh, Srijan Mishra, Thomas Foubert, Abhinav Rastogi, Adam Yang, Albert Q. Jiang, Alexandre Sablayrolles, Amélie Héliou, and 87 others. 2025. Voxtral. Preprint, arXiv:2507.13264. [↗](https://arxiv.org/abs/2507.13264)
19. Sewon Min, Xinxi Lyu, Ari Holtzman, Mikel Artetxe, Mike Lewis, Hannaneh Hajishirzi, and Luke Zettlemoyer. 2022. [Rethinking the role of demonstrations: What makes in-context learning work?](https://md.research.elamur.ai/free/en/papers/min-2022-rethinking-role-demonstrations-what-makes/) In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 11048–11064, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics.
20. Tobi Olatunji, Tejumade Afonja, Aditya Yadavalli, Chris Chinenye Emezue, Sahib Singh, Bonaventure F. P. Dossou, and 1 others. 2023. [Afrispeech-200: Pan-african accented speech dataset for clinical and general domain asr](https://md.research.elamur.ai/free/en/papers/olatunji-2023-afrispeech-200-pan-african-accented/). Transactions of the Association for Computational Linguistics, 11:1599–1617.
21. Vassil Panayotov, Guoguo Chen, Daniel Povey, and Sanjeev Khudanpur. 2015. Librispeech: An ASR corpus based on public domain audio books. In Proceedings of the 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (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+the+2015+IEEE+International+Conference+on+Aco)
22. Krishna C. Puvvada, Piotr Żelasko, He Huang, Oleksii Hrinchuk, Nithin Rao Koluguri, Kunal Dhawan, Somshubra Majumdar, Elena Rastorgueva, Zhehuai Chen, Vitaly Lavrukhin, Jagadeesh Balam, and Boris Ginsburg. 2024. Less is more: Accurate speech recognition & translation without web-scale data. arXiv preprint arXiv:2406.19674. [↗](https://arxiv.org/abs/2406.19674)
23. Ewout W. Steyerberg. 2021. The digital scribe in clinical practice: A scoping review and research agenda. npj Digital Medicine, 4:57. [find](https://scholar.google.com/scholar?q=Ewout+W%2E+Steyerberg%2E+2021%2E+The+digital+scribe+in+clinical+practice%3A+A+scoping+review+and+research+agenda%2E+npj+Digital+Medicine%2C+4%3A57%2E)
24. Lingli Wang, Ni Huang, Yili Hong, Luning Liu, Xun-hua Guo, and Guoqing Chen. 2023. Voice-based AI in call center customer service: A natural field experiment. Production and Operations Management, 32(4):1002–1018. [↗](https://doi.org/10.1111/poms.13953)
25. Weiyi Wang, Chau Tran, Fahim Azhar, Henrik Rottmann, Armand Joulin, and 1 others. 2021. Voxpopuli: A large-scale multilingual speech corpus for representation, semi-, and self-supervised learning. In Proceedings of Interspeech 2021, pages 993–997. [find](https://scholar.google.com/scholar?q=Weiyi+Wang%2C+Chau+Tran%2C+Fahim+Azhar%2C+Henrik+Rottmann%2C+Armand+Joulin%2C+and+1+others%2E+2021%2E+Voxpopuli%3A+A+large%2Dscale+multilingual+speech+corpus+for+representation%2C+semi%2D%2C+and+self%2Dsupervised+learning%2E+In+)
26. Mario Zusag, Laurin Wagner, and Bernhad Thallinger. 2024. Crisperwhisper: Accurate timestamps on verbatim speech transcriptions. In Interspeech 2024, pages 1265–1269. [↗](https://doi.org/10.48550/arxiv.2408.16589)
27. 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 the 40th International Conference on Machine Learning, volume 202 of Proceedings of Machine Learning Research, pages 28492–28518. PMLR. [↗](https://doi.org/10.48550/arxiv.2212.04356)
28. Dima Rekesh, Nithin Rao Koluguri, Samuel Kriman, Somshubra Majumdar, Vahid Noroozi, He Huang, Oleksii Hrinchuk, Krishna C. Puvvada, Ankur Kumar, Jagadeesh Balam, and Boris Ginsburg. 2023. Fast conformer with linearly scalable attention for efficient speech recognition. arXiv preprint arXiv:2305.05084. [↗](https://arxiv.org/abs/2305.05084)
29. Hadeel Saadany, Catherine Breslin, Constantin Orasan, and Sophie Walker. 2023. Better transcription of uk supreme court hearings. In AI4AJ@ICAIL. [↗](https://doi.org/10.48550/arxiv.2211.17094)
30. Mardhiyah Sanni, Tassallah Abdullahi, Devendra Kayande, Emmanuel Ayodele, Naome Etori, Michael Mollel, and 1 others. 2025. [Afrispeech-dialog: A benchmark dataset for spontaneous english conversations in healthcare and beyond](https://md.research.elamur.ai/free/en/papers/sanni-2025-afrispeech-dialog-benchmark-dataset-spontaneous/). In Proceedings of the 2025 Conference of the North American Chapter of the Association for Computational Linguistics (NAACL).
31. Holger Schwenk, Loïc Barrault, Yu-An Chung, Francisco Guzmán, Juan Pino, and the Seamless Communication Team. 2023. Seamlessm4t: Massively multilingual and multimodal machine translation. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP). [find](https://scholar.google.com/scholar?q=Holger+Schwenk%2C+Lo%C3%AFc+Barrault%2C+Yu%2DAn+Chung%2C+Francisco+Guzm%C3%A1n%2C+Juan+Pino%2C+and+the+Seamless+Communication+Team%2E+2023%2E+Seamlessm4t%3A+Massively+multilingual+and+multimodal+machine+translation%2E+In+Proceedin)

*Updated 2 November 2026*