Elamur Research⌕
☰
home / papers / 2026 / BIPA: Brazilian Portuguese Phonetic Dataset with Dialectal Variations in IPA Standard

BIPA: Brazilian Portuguese Phonetic Dataset with Dialectal Variations in IPA Standard

Paper · 2026 · Thiago Monteles de Sousa, Lucas Rafael Gris and Nádia Félix Felipe da Silva

Summary

Brazilian Portuguese has few open phonetic resources, and existing grapheme-to-phoneme (G2P) dictionaries mostly cover only a single standard pronunciation and ignore regional accents. The authors built BIPA, a pronunciation corpus scraped automatically from Wiktionary pages carrying dialect-labelled IPA transcriptions. Their pipeline has four stages: collection via HTML parsing and regular expressions, normalization of IPA symbols and dialect tags, filtering, and export to JSON Lines. The filtering stage removes acronyms, most proper names and entries with non-native sounds. In the resulting corpus, every word is linked to transcriptions in six varieties: the Northeast, the Center-West, the South, Rio de Janeiro, São Paulo, and a general Brazilian standard. To show the data is usable, they fine-tuned ByT5-small as a dialect-aware G2P model. Its best minimum phoneme error rate was 3.6%, and it clearly beat the multilingual CharsiuG2P model on standard Brazilian Portuguese. Accuracy drops sharply for dialects with little data, which exposes the corpus's regional imbalance. Both the dataset and the model are public on Hugging Face. They can support accent-aware TTS, ASR and sociolinguistic research.

Key points

  • BIPA holds 53,353 unique words (46,518 lemmas) and 350,021 IPA transcriptions, about 6.56 per word. The data was taken from Wiktionary on 17 Sept 2025 under CC BY-SA 4.0. For comparison, WikiPron covers only 9,315 Brazilian Portuguese pairs.
  • ByT5-small was trained for 10 epochs and reached a best Minimum PER of 3.6%.
  • Per-dialect validation Min PER:
    • São Paulo: 1.93%
    • South: 2.36%
    • Rio de Janeiro: 3.08%
    • general Brazil: 6.07%
    • Center-West: 42.86% (only 7 examples)
    • Northeast: 78.12% (only 32 examples)
  • On the standard dialect, BIPA-ByT5 scored 6.07% against 29.32% for CharsiuG2P (Zhu et al., 2022). With stress and syllable marks stripped, the scores were 4.81% and 6.40%.
  • For minority dialects, the model tends to fall back to "Brazil (general)" pronunciations, for example with mainha and inteiro.

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 2The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 works

Cite

Cite as Thiago Monteles de Sousa, Lucas Rafael Gris and Nádia Félix Felipe da Silva. 2026. BIPA: Brazilian Portuguese Phonetic Dataset with Dialectal Variations in IPA Standard. Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1.

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

@inproceedings{sousa2026bipa,
  title = {{BIPA: Brazilian Portuguese Phonetic Dataset with Dialectal Variations in IPA Standard}},
  author = {Thiago Monteles {de Sousa}, Lucas Rafael Gris and Nádia Félix Felipe {da Silva}},
  booktitle = {Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 1},
  year = {2026},
  url = {https://aclanthology.org/2026.propor-1.47/}
}

References

38 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Rúben Almeida, Ricardo Campos, Alípio Jorge, and Sérgio Nunes. 2024. Indexing portuguese nlp resources with pt-pump-up. In Proceedings of the 16th International Conference on Computational Processing of Portuguese, pages 178–181. find
  2. Amadeu Amaral. 1920. O dialeto caipira. Casa Editora O Livro, São Paulo, SP. find
  3. Lu Dong, Zhi-Qiang Guo, Chao-Hong Tan, Ya-Jun Hu, Yuan Jiang, and Zhen-Hua Ling. 2022. Neural grapheme-to-phoneme conversion with pre-trained grapheme models. Preprint, arXiv:2201.10716. — [arXiv:2201.10716](https://arxiv.org/abs/2201.10716) find
  4. David M. Eberhard, Gary F. Simons, and Charles D. Fennig. 2025. Ethnologue: Languages of the world. find
  5. Rodrigo Garcia Garay. 2016. O fonema: linguística e história. find
  6. International Phonetic Association. 1999. Handbook of the International Phonetic Association: A Guide to the Use of the International Phonetic Alphabet. Cambridge University Press. find
  7. Plínio A. Barbosa and Eleonora C. Albano. 2004. Brazilian Portuguese. Journal of the International Phonetic Association, 34(2):227–232. find
  8. Elisa Battisti. 2017. Harmonia vocálica de altura no português brasileiro em formas nominais não derivadas: análise de um processo variável pela teoria da otimi-dade. find
  9. Claudia Brescancini and Valéria Neto de Oliveira Monaretto. 2008. Os róticos no sul do brasil: panorama e generalizações. Signum: Estudos da Linguagem, 11(2):51–66. find
  10. Zébulon Goriely and Paula Buttery. 2025. Ipa-childes g2p+: Feature-rich resources for cross-lingual phonology and phonemic language modeling. Preprint, arXiv:2504.03036.
  11. Henry Grafé and 1 others. 2025. Graph connectionist temporal classification for phoneme recognition. arXiv preprint arXiv:2509.05399. — [arXiv:2509.05399](https://arxiv.org/abs/2509.05399) find
  12. Alex Graves. 2012. Sequence transduction with recurrent neural networks. In Representation Learning Workshop, ICML.
  13. International Phonetic Association. 2015. Ipa chart. 107 segmental letters, 44 diacritics, 4 prosodic marks. find
  14. Edresson Casanova, Vinícius Gonçalves dos Santos, Flaviane Romani Fernandes Svartman, Marli Quadros Leite, Arnaldo Candido Junior, Ricardo Marcondes Marcacini, Solange Oliveira Rezende, and Sandra Maria Aluísio. 2023. Recursos para o processamento de fala. Processamento de linguagem natural: conceit… find
  15. TaRSila Project Contributors. 2024. Tarsila: Growing speech datasets for brazilian portuguese. find
  16. Nicholas Kluge Corrêa, Aniket Sen, and 1 others. 2024. Tucano: Advancing neural text generation for portuguese. Patterns. University of Bonn. find
  17. William Alberto Cruz-Castañeda and Marcellus Amadeus. 2025. Large languages models in brazilian portuguese: A chronological survey. Journal of the Brazilian Computer Society, 31(1):1168–1187. find
  18. Júlia da Rocha Junqueira, Émerson Lopes, Larissa Freitas, and Ulisses B Correa. 2024. A systematic analysis of multilingual and low-resource languages models: A review on brazilian portuguese. find
  19. Márcia Santos Duarte de Oliveira and Maria de Lurdes Zanolli. 2021. O /r/ retroflexo no português caipira como resultado de "interferência" da língua geral de são paulo: uma homenagem à obra de amadeu amaral. Estudos Linguísticos, 50(3):1159– 1172. find
  20. Aliya Deri and Kevin Knight. 2016. Grapheme-to-phoneme models for (almost) any language. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 399–408. find
  21. Michael Kenstowicz and Filomena Sandalo. 2016. Pretonic vowel reduction in brazilian portuguese: Harmony and dispersion. Journal of Portuguese linguistics, 15(1). find
  22. Hamza Kheddar, Mustapha Hemis, and Yassine Himeur. 2024. Automatic speech recognition using advanced deep learning approaches: A survey. Information fusion, 109:102422. find
  23. Hwa-Yeon Kim, Jong-Hwan Kim, and Jae-Min Kim. 2022. Fast bilingual grapheme-to-phoneme conversion. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies: Industry Track, pages 289–296, Hybrid: Seattle, Washin… find
  24. Jackson L. Lee, Lucas F.E. Ashby, M. Elizabeth Garza, Yeonju Lee-Sikka, Sean Miller, Alan Wong, Arya D. McCarthy, and Kyle Gorman. 2020. Massively multilingual pronunciation modeling with WikiPron. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 4223–4228, Marseille… find
  25. Vladimir I Levenshtein. 1966. Binary codes capable of correcting deletions, insertions, and reversals. Soviet physics doklady, 10(8):707–710. find
  26. Rodrigo Lima, Sidney E Leal, Arnaldo Candido Junior, and Sandra M Aluísio. 2024. A large dataset of spontaneous speech with the accent spoken in são paulo for automatic speech recognition evaluation. In Brazilian Conference on Intelligent Systems, pages 33–47. Springer. find
  27. Gustavo Mendonça and Sandra M Aluísio. 2014. Using a hybrid approach to build a pronunciation dictionary for brazilian portuguese. In INTERSPEECH, pages 1278–1282. find
  28. Steven Moran and Daniel McCloy, editors. 2019. PHOIBLE 2.0. Max Planck Institute for the Science of Human History, Jena. find
  29. Ben Peters, Jon Dehdari, and Josef van Genabith. 2017. Massively multilingual neural grapheme-to-phoneme conversion. In Proceedings of the First Workshop on Building Linguistically Generalizable NLP Systems, pages 19–26, Copenhagen, Denmark. Association for Computational Linguistics. find
  30. Noam Shazeer and Mitchell Stern. 2018. Adafactor: Adaptive learning rates with sublinear memory cost. arXiv preprint arXiv:1804.04235. find
  31. Luann Dias Souza. 2025. Acento e peso silábico no português brasileiro. find
  32. Xu Tan, Tao Qin, Frank Soong, and Tie-Yan Liu. 2021. A survey on neural speech synthesis. arXiv preprint arXiv:2106.15561. find
  33. Siddharth A Unnithan and Nihar R Mahapatra. 2024. Advancements in grapheme-to-phoneme conversion models for speech synthesis. In International Conference on Computational Science and Computational Intelligence, pages 139–150. Springer. find
  34. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in neural information processing systems, volume 30, pages 5998–6008.
  35. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier-ric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, and 3 others. 2020. Transf…
  36. Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, and Colin Raffel. 2022. Byt5: Towards a token-free future with pre-trained byte-to-byte models. Transactions of the Association for Computational Linguistics, 10:291–306. find
  37. Jian Zhu, Cong Zhang, and David Jurgens. 2022. Byt5 model for massively multilingual grapheme-to-phoneme conversion. Preprint, arXiv:2204.03067. — [arXiv:2204.03067](https://arxiv.org/abs/2204.03067) find
  38. Markéta Řezáčková, Jan Švec, and Daniel Tihelka. 2021. T5g2p: Using text-to-text transfer transformer for grapheme-to-phoneme conversion. In Interspeech 2021, pages 6–10. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sousa-2026-bipa-brazilian-portuguese-phonetic-dataset/