Elamur Research⌕
☰
home / papers / 2026 / Toward Equitable Machine Translation for Atypical Speech: An LLM Post-Correction Approach

Toward Equitable Machine Translation for Atypical Speech: An LLM Post-Correction Approach

Summary

Speech recognition makes far more mistakes for people with speech disorders. In cascaded speech translation, those mistakes carry over into the translated output. The authors trace this chain for two conditions. The first is real dysarthric speech from the Torgo corpus, transcribed with a large Whisper model, which gives 691 sentence-level lines. The second is rhotacism, which they simulate by substituting rhotic minimal-pair words into Tatoeba sentences, since suitable real recordings are scarce. They then test whether open-source LLMs, used without any training as transcript post-correctors, can repair the damage before translation. Translation is evaluated into three languages chosen to span resource availability: Spanish as the high-resource case, Ukrainian as the medium one, and Khmer as the low one. Gains at the recognition level are small. Even so, the two largest models, LLaMA3 70B and Qwen3 235B, consistently improve translation quality in all three directions, while OLMo-3 7B does not. The study points to a cheap, plug-in route toward fairer speech technology, though it depends on model size. Ukrainian is one of the evaluated targets, and the code and data are public.

Key points

  • Data: dysarthric sentences come from Torgo (head-mounted microphone, control speakers removed, transcribed by Whisper large). Simulated rhotacism comes from Tatoeba parallel sentences filtered for rhotic content.
  • Correctors: OLMo-3 (7B), LLaMA3 (70B) and Qwen3 (235B), each driven by one fixed prompt per condition with no fine-tuning.
  • WER results: relative WER improvement reaches at most about 1.4% for dysarthria and up to about 2.6% for rhotacism.
  • Translation results: LLaMA3 and Qwen3 improve translation quality across EN→ES/UK/KM. Qwen3-corrected rhotacism output gets close to the quality of clean-input translation. OLMo-3 7B shows no reliable gains.
  • Limitations: Torgo has no reference translations, so its outputs are scored only with the reference-free metrics XCOMET and MetricX. Tatoeba may overlap with model training data, and the rhotacism condition is text-simulated rather than recorded. Code and data are at github.com/byu-matrix-lab/project-atypical-speech-mt.

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: Linguistics & languages · Machine translation

Topics: Machine translation

Cite

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

@inproceedings{pasion2026toward,
  title = {{Toward Equitable Machine Translation for Atypical Speech: An LLM Post-Correction Approach}},
  author = {Pasion, Grace and Shurtz, Ammon and Richardson, Stephen D.},
  booktitle = {Proceedings of the 17th Conference of the Association for Machine Translation in the Americas (Volume 1: Research Track)},
  year = {2026},
  url = {https://aclanthology.org/2026.amta-research.3/}
}

References

54 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ahmed Aboeitta, Ahmed Sharshar, Youssef Nafea, and Shady Shehata. Bridging asr and llms for dysarthric speech recognition: Benchmarking self-supervised and generative approaches. arXiv preprint arXiv:2508.08027, 2025. — [arXiv:2508.08027](https://arxiv.org/abs/2508.08027) find
  2. Chitralekha Bhat and Helmer Strik. Two-stage data augmentation for improved asr performance for dysarthric speech. Computers in Biology and Medicine, 189: 109954, 2025. find
  3. Peter Flipsen Jr. Remediation of/r/for speech-language pathologists, volume 1. Plural Publishing, 2021. find
  4. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783, 2024. find
  5. Nuno M Guerreiro, Ricardo Rei, Daan van Stigt, Luisa Coheur, Pierre Colombo, and André FT Martins. xcomet: Proceedings of the 17th Conference of the Association for Machine Translation in the Americas, Québec City, Canada, August 31 - September 2, 2026. Volume 1: Research Track Transparent machine… find
  6. Mark Hasegawa-Johnson, Xiuwen Zheng, Heejin Kim, Clarion Mendes, Meg Dickinson, Erik Hege, Chris Zwilling, Marie Moore Channell, Laura Mattie, Heather Hodges, et al. Community-supported shared infrastructure in support of speech accessibility. Journal of Speech, Language, and Hearing Research, 67(11… find
  7. Yongjun He, Guanglu Sun, and Jiqing Han. Spectrum enhancement with sparse coding for robust speech recognition. Digital Signal Processing, 43:59–70, 2015. find
  8. Abner Hernandez, Paula Andrea Pérez-Toro, Elmar Nöth, Juan Rafael Orozco-Arroyave, Andreas Maier, and Seung Hee Yang. Cross-lingual self-supervised speech representations for improved dysarthric speech recognition, 2022. URL https://arxiv.org/abs/2204.01670. — [arXiv:2204.01670](https://arxiv.org/abs/2204.01670) ↗
  9. Abner Hernandez, Tomás Arias-Vergara, Andreas Maier, and Paula Andrea Pérez-Toro. Enhancing asr accuracy for speakers with parkinson’s disease using instruction-tuned llms. In International Conference on Text, Speech, and Dialogue, pages 153–164. Springer, 2025a. find
  10. Abner Hernandez, Tomás Arias Vergara, Andreas Maier, and Paula Andrea Pérez-Toro. Confidence-guided error correction for disordered speech recognition. arXiv preprint arXiv:2509.25048, 2025b. — [arXiv:2509.25048](https://arxiv.org/abs/2509.25048) find
  11. Yu-Chen Hsu and Yue-Shan Chang. A dysarthric speech recognition system with personal style embedding. In 2025 IEEE International Conference on Systems, Man, and Cybernetics (SMC), pages 3770–3775. find
  12. Yuchen Hu, Chen Chen, Chao-Han Yang, Chengwei Qin, Pin-Yu Chen, Eng-Siong Chng, and Chao Zhang. Self-taught recognizer: Toward unsupervised adaptation for speech foundation models. Advances in Neural Information Processing Systems, 37:29566–29594, 2024. find
  13. Shangkun Huang, Jing Deng, Jintao Kang, and Rong Zheng. Leveraging llm for stuttering speech: A unified architecture bridging recognition and event detection. arXiv preprint arXiv:2505.22005, 2025. — [arXiv:2505.22005](https://arxiv.org/abs/2505.22005) find
  14. Macarious Hui, Jinda Zhang, and Aanchan Mohan. Enhancing aac software for dysarthric speakers in e-health settings: an evaluation using torgo. In ICC 2025-IEEE International Conference on Communications, pages 3673–3679. find
  15. Artyom Iudin, Dmitrii Korzh, Matvey Skripkin, Dmitrii Masnyi, and Oleg Rogov. Enhancing speech recognition with llms in post-correction settings. Madeline Jefferson. Usability of automatic speech recognition systems for individuals with speech disorders: Past, present, future, and a proposed model.… find
  16. Juraj Juraska, Daniel Deutsch, Mara Finkelstein, and Markus Freitag. MetricX-24: The Google submission to the WMT 2024 metrics shared task. In Barry Haddow, Tom Kocmi, Philipp Koehn, and Christof Monz, editors, Proceedings of the Ninth Conference on Machine Translation, pages 492–504, Miami, Florida… find our card
  17. Juraj Juraska, Tobias Domhan, Mara Finkelstein, Tetsuji Nakagawa, Geza Kovacs, Daniel Deutsch, Pidong Wang, and Markus Freitag. Metricx-25 and gemspaneval: Google translate submissions to the wmt25 evaluation shared task. In Proceedings of the Tenth Conference on Machine Translation, pages 957–968,… find
  18. Gautam Krishna, Mason Carnahan, Shilpa Shamapant, Yashitha Surendranath, Saumya Jain, Arundhati Ghosh, Co Tran, Jose Del R Millan, and Ahmed H Tewfik. Brain signals to rescue aphasia, apraxia and dysarthria speech recognition. In 2021 43rd Annual International Conference of the IEEE Engineering in M… find
  19. Moreno La Quatra, Alkis Koudounas, Valerio Mario Salerno, and Sabato Marco Siniscalchi. Exploring generative error correction for dysarthric speech recognition. arXiv preprint arXiv:2505.20163, 2025. — [arXiv:2505.20163](https://arxiv.org/abs/2505.20163) find
  20. Wonjun Lee, San Kim, and Gary Geunbae Lee. Enhancing dialogue speech recognition with robust contextual awareness via noise representation learning. In Tatsuya Kawahara, Vera Demberg, Stefan Ultes, Koji Inoue, Shikib Mehri, David Howcroft, and Kazunori Komatani, editors, Proceedings of the 25th Annu… find
  21. Proceedings of the 17th Conference of the Association for Machine Translation in the Americas, Québec City, Canada, August 31 - September 2, 2026. Volume 1: Research Track Haoyang Li, Changsong Liu, Wei Rao, Hao Shi, Sakriani Sakti, and Eng Siong Chng. Training-free intelligibility-guided observatio… — [arXiv:2602.20967](https://arxiv.org/abs/2602.20967) find
  22. Zhaolin Li, Yining Liu, Danni Liu, Tuan Nam Nguyen, Enes Yavuz Ugan, Tu Anh Dinh, Carlos Mullov, Alexander Waibel, and Jan Niehues. KIT’s low-resource speech translation systems for IWSLT2025: System enhancement with synthetic data and model regularization. In Elizabeth Salesky, Marcello Federico, a… find
  23. Ye Bhone Lin, Thura Aung, Ye Kyaw Thu, and Thazin Myint Oo. Asr error correction in low-resource burmese with alignment-enhanced transformers using phonetic features. In 2025 20th International Joint Symposium on Artificial Intelligence and Natural Language Processing (iSAI-NLP), pages 1–6. find
  24. Haowei Lou, Chengkai Huang, Hye-young Paik, Yongquan Hu, Aaron Quigley, Wen Hu, and Lina Yao. Speecha-gent: An end-to-end mobile infrastructure for speech impairment assistance. arXiv preprint arXiv:2510.20113, 2025. — [arXiv:2510.20113](https://arxiv.org/abs/2510.20113) find
  25. Rao Ma, Mengjie Qian, Mark Gales, and Kate Knill. Asr error correction using large language models. IEEE Transactions on Audio, Speech and Language Processing, 2025. find
  26. Hari Krishna Maganti and Marco Matassoni. Auditory processing-based features for improving speech recognition in adverse acoustic conditions. EURASIP Journal on Audio, Speech, and Music Processing, 2014(1):21, 2014. find
  27. Anirudh Mani, Shruti Palaskar, Nimshi Venkat Meripo, Sandeep Konam, and Florian Metze. Asr error correction and domain adaptation using machine translation. In ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 6344–6348. find
  28. Marco Marini, Nicola Vanello, and Luca Fanucci. Optimising speaker-dependent feature extraction parameters to improve automatic speech recognition performance for people with dysarthria. Sensors, 21(19):6460, 2021. find
  29. Andres Marzal and Enrique Vidal. Computation of normalized edit distance and applications. IEEE transactions on pattern analysis and machine intelligence, 15(9): 926–932, 2002. find
  30. E Matusov, S Kanthak, and Hermann Ney. On the integration of speech recognition and statistical machine translation. In Proc. Interspeech 2005, pages 3177–3180, 2005. find
  31. Kinfe Tadesse Mengistu and Frank Rudzicz. Adapting acoustic and lexical models to dysarthric speech. In 2011 IEEE international conference on acoustics, speech and signal processing (ICASSP), pages 4924–4927. find
  32. Bernd T Meyer, Thomas Brand, and Birger Kollmeier. Effect of speech-intrinsic variations on human and automatic recognition of spoken phonemes. The Journal of the Acoustical Society of America, 129(1):388–403, 2011. find
  33. Anna Min, Chenxu Hu, Yi Ren, and Hang Zhao. When end-to-end is overkill: Rethinking cascaded speech-to-text translation. arXiv preprint arXiv:2502.00377, 2025. — [arXiv:2502.00377](https://arxiv.org/abs/2502.00377) find
  34. Megan A Morris, Sarah K Meier, Joan M Griffin, Megan E Branda, and Sean M Phelan. Prevalence and etiologies of adult communication disabilities in the united states: Results from the 2012 national health interview survey. Disability and health journal, 9(1):140–144, 2016. find
  35. Denis Moser, Nikola Stanic, and Murat Sariyar. Benchmarking speech-to-text robustness in noisy emergency medical dialogues: an evaluation of models under realistic acoustic conditions. JAMIA open, 8(6):ooaf147, 2025. find
  36. Ramy Mounir, Redwan Alqasemi, and Rajiv Dubey. Speech assistance for persons with speech impediments using artificial neural networks. In ASME International Mechanical Engineering Congress and Exposition, volume 58363, page V003T04A056. American Society of Mechanical Engineers, 2017. find
  37. Team Olmo, Allyson Ettinger, Amanda Bertsch, Bailey Kuehl, David Graham, David Heineman, Dirk Groeneveld, Faeze Brahman, Finbarr Timbers, Hamish Ivison, Jacob Morrison, Jake Poznanski, Kyle Lo, Luca Soldaini, Matt Jordan, Mayee Chen, Michael Noukhovitch, Proceedings of the 17th Conference of the Ass… find
  38. Matt Post. A call for clarity in reporting bleu scores. In Proceedings of the third conference on machine translation: Research papers, pages 186–191, 2018. find our card
  39. Jie Pu, Thai-Son Nguyen, and Sebastian Stüker. Multi-stage large language model correction for speech recognition. arXiv preprint arXiv:2310.11532, 2023. — [arXiv:2310.11532](https://arxiv.org/abs/2310.11532) find
  40. Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. Robust speech recognition via large-scale weak supervision. In International conference on machine learning, pages 28492– 28518. PMLR, 2023. find our card
  41. Frank Rudzicz, Aravind Kumar Namasivayam, and Talya Wolff. The torgo database of acoustic and articulatory speech from speakers with dysarthria. Language resources and evaluation, 46(4):523–541, 2012. find
  42. Megha Rughani and D. Shivakrishna. Hybridized feature extraction and acoustic modelling approach for dysarthric speech recognition, 2015. URL https://arxiv.org/abs/1506.02170. dysarthric speech recognition system. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 31:3407–3416, 202… — [arXiv:1506.02170](https://arxiv.org/abs/1506.02170) ↗
  43. Mohammed Sidi Yakoub, Sid-ahmed Selouani, Brahim-Fares Zaidi, and Asma Bouchair. Improving dysarthric speech recognition using empirical mode decomposition and convolutional neural network. EURASIP Journal on Audio, Speech, and Music Processing, 2020(1):1, 2020. find
  44. Zhiyuan Tang, Dong Wang, Shen Huang, and Shidong Shang. Full-text error correction for chinese speech recognition with large language model. In ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5. find
  45. Jörg Tiedemann. The tatoeba translation challenge– realistic data sets for low resource and multilingual mt. arXiv preprint arXiv:2010.06354, 2020. find our card
  46. Jimmy Tobin, Phillip Nelson, Bob MacDonald, Rus Heywood, Richard Cave, Katie Seaver, Antoine Desjardins, Pan-Pan Jiang, and Jordan R Green. Automatic speech recognition of conversational speech in individuals with disordered speech. Journal of Speech, Language, and Hearing Research, 67(11):4176–4185… find
  47. Zhong-Qiu Wang, Peidong Wang, and DeLiang Wang. Complex spectral mapping for single-and multi-channel speech enhancement and robust asr. IEEE/ACM transactions on audio, speech, and language processing, 28: 1778–1787, 2020. find
  48. Felix Weninger, Hakan Erdogan, Shinji Watanabe, Emmanuel Vincent, Jonathan Le Roux, John R Hershey, and Björn Schuller. Speech enhancement with lstm recurrent neural networks and its application to noise-robust asr. In International conference on latent variable analysis and signal separation, page… find
  49. Paban Sapkota, Hemant Kumar Kathania, Shrikanth Narayanan, and Sudarsana Reddy Kadiri. Addressing dysarthric speech variability with severity-aware fine-tuning of transformer-based wav2vec2 asr models: P. sapkota, hk kathania, s. narayanan, sr kadiri. Circuits, Systems, and Signal Processing, pages… find
  50. Cihan Xiao, Matthew Wiesner, Debashish Chakraborty, Reno Kriz, Keith Cunningham, Kenton Murray, Kevin Duh, Luis Tavarez-Arce, Paul McNamee, and Sanjeev Khudanpur. Whisper-ut: A unified translation framework for speech and text. In Proceedings of the 2025 Conference on Empirical Methods in Natural La… find
  51. Seyed Reza Shahamiri, Vanshika Lal, and Dhvani Shah. Dysarthric speech transformer: A sequence-to-sequence An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chen-Proceedings of the 17th Conference of the Association for Machine Translation in the Americas,… find our card
  52. Chao-Han Huck Yang, Yile Gu, Yi-Chieh Liu, Shalini Ghosh, Ivan Bulyko, and Andreas Stolcke. Generative speech recognition error correction with large language models and task-activating prompting. In 2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), pages 1–8. find
  53. Xiaohu Zhao, Haoran Sun, Yikun Lei, and Deyi Xiong. Regularizing cross-attention learning for end-to-end speech translation with asr and mt attention matrices. Expert Systems with Applications, 247:123241, 2024. find
  54. Proceedings of the 17th Conference of the Association for Machine Translation in the Americas, Québec City, Canada, August 31 - September 2, 2026. Volume 1: Research Track find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/pasion-2026-equitable-machine-translation-atypical-speech/