Elamur Research⌕
☰
home / papers / 2026 / Building Collaborative Speech Corpora for Low-Resource Languages: The Galician Dataset in Mozilla Common Voice

Building Collaborative Speech Corpora for Low-Resource Languages: The Galician Dataset in Mozilla Common Voice

Summary

Galician is a low-to-medium resource language. Its text resources are growing, but it still lacks open speech data with enough speaker and accent variety. Several existing ASR corpora are also tied to narrow domains, such as parliamentary sessions or spoken language understanding. The authors, from the Instituto da Lingua Galega and working within Proxecto Nós, describe how the Galician portion of Mozilla Common Voice (MCV) was assembled. They draw on two precedents: the FalAI crowdsourcing campaign and the Catalan AINA initiative. Their own campaigns combined in-person recording events, student involvement, Validatón validation marathons and partnerships with companies. Examining the Galician data in MCV v22.0, they find that the corpus has reached a moderate size. However, speaker metadata are often incomplete and dialect labels are scarce, which reduces its value for ASR. The paper concludes that clear messaging about data use, culturally familiar prompts and simple tools encourage participation. Representativeness, contributor retention and ethical governance, by contrast, require deliberate planning. These lessons may help other under-resourced language communities.

Key points

  • MCV releases all data under CC0, in line with FAIR principles. Each clip needs approval from two independent listeners, and the local community curates the text prompts.
  • The Galician analysis is based on MCV v22.0. The main weaknesses are missing speaker metadata and limited dialect tagging, both of which affect ASR performance.
  • Other open Galician speech resources:
    • Nos_ParlaSpeech-GL: about 1,600 h of parliamentary speech.
    • Nos_TranscriSpeech-GL: 53 h across multiple domains.
    • FalAI: 250 h.
    • OpenSLR77: about 10 h.
    • FLEURS: Galician subset.
    • TTS datasets: Nos_Celtia-GL, Nos_Brais-GL and CRPIH_UVigo-GL-Voices (about 21 h).
  • Reference campaigns:
    • FalAI drew more than 11,000 volunteers and over 250 h of speech, against a 100 h goal.
    • Catalan AINA exceeded 3,500 h in MCV, supported by paid validation rounds.
  • Proxecto Nós is led by ILG and CiTIUS at USC, with backing from the Xunta de Galicia. It operates within the state-level ILENIA and ALIA projects.

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

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/4gd79cq3cump

@inproceedings{vladu2026building,
  title = {{Building Collaborative Speech Corpora for Low-Resource Languages: The Galician Dataset in Mozilla Common Voice}},
  author = {Vladu, Adina Ioana and Rei, Elisa Fernández and Lago, María Pérez},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/4gd79cq3cump},
  url = {https://doi.org/10.63317/4gd79cq3cump}
}

References

18 references, 1 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 Henretty, Michael Kohler, Josh Meyer, Reuben Morais, Lindsay Saunders, Francis M. Tyers, and Gregor Weber. 2020. Common Voice: A Massively-Multilingual Speech Corpus. find our card
  2. Carme Armentano-Oller, Montserrat Marimon, and Marta Villegas. 2024. Becoming a High-Resource Language in Speech: The Catalan Case in the Common Voice Corpus. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024),… find
  3. Eduardo Rodríguez Banga, Carmen García-Mateo, Francisco Méndez-Pazó, Manuel González-González, and Carmen Magariños. 2012. Cotovía: an open source TTS for Galician and Spanish. In VII Jornadas en Tecnología del Habla and III Iberian SLTech Workshop, IberSPEECH, pages 308–315. find
  4. Iria de Dios-Flores, Carmen Magariños, Adina Ioana Vladu, John E. Ortega, José Ramom Pichel, Marcos García, Pablo Gamallo, Elisa Fernández Rei, Alberto Bugarín-Diz, Manuel González González, Senén Barro, and Xosé Luis Regueira. 2022. The Nós Project: Opening routes for the Galician language in the f… find
  5. Jan Krewer. 2023. Creating Community-Driven Datasets: Insights from Mozilla Common Voice Activities in East Africa. Technical report, Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) GmbH, Bonn, Germany. find
  6. Baybars Külebi, Inma Hernáez, Elisa Fernández Rei, Andres Montoyo, Sarah Solito, Carme Armentano-Oller, Javier Hernando, Eva Navas, Carmen Magariños, Adina Vladu, Ibon Saratxaga, Jon Sánchez, Victor García Romillo, Asier Herranz, Christoforos Souganidis, Noelia García, Antonio Moscoso Sánchez, Xose… find
  7. Andres Pineiro-Martin, Carmen Garcia-Mateo, Laura Docio-Fernandez, Maria del Carmen Lopez-Perez, and Jose Gandarela-Rodriguez. 2024. FalAI: A dataset for end-to-end spoken language understanding in a low-resource scenario. In Proc. of the 2024 Joint Int. Conf. on Comput. find
  8. Ling., Lang. Resour. and Eval. (LREC-COLING 2024), pages 7107–7116, Torino, Italy. ELRA and ICCL. find
  9. José Manuel Ramírez Sánchez and Carmen García-Mateo. 2022. Report on the Galician Language. Report D1.15, ELE. find
  10. Adina Ioana Vladu, Iria de Dios-Flores, Carmen Magariños, John E. Ortega, José Ramom Pichel, Marcos Garcia, Pablo Gamallo, Elisa Fernández Rei, Alberto Bugarín, Manuel González González, Senén Barro, and Xosé Luis Regueira. 2022. Proxecto Nós: Artificial intelligence at the service of the Galician l… find
  11. Changhan Wang, Juan Pino, Anne Wu, and Jiatao Gu. 2020a. CoVoST: A diverse multilingual speech-to-text translation corpus. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 4197–4203, Marseille, France. European Language Resources Association. find
  12. Changhan Wang, Anne Wu, and Juan Miguel Pino. 2020b. Covost 2: A massively multilingual speech-to-text translation corpus. CoRR, abs/2007.10310. find
  13. Mark D Wilkinson, Michel Dumontier, Ijsbrand Jan Aalbersberg, Gabrielle Appleton, Myles Axton, Arie Baak, Niklas Blomberg, Jan-Willem Boiten, Luiz Bonino da Silva Santos, Philip E Bourne, Jildau Bouwman, Anthony J Brookes, Tim Clark, Mercè Crosas, Ingrid Dillo, Olivier Dumon, Scott Edmunds, Chris T… find
  14. Language Resource References Magariños, Carmen and Adrián Vidal Miguéns and Adina Ioana Vladu and Noelia García Díaz and Marta Vázquez Abuín and Ainhoa Vivel Couso and Daniel Bardanca and Elisa Fernández Rei. 2023. Nos_ParlaSpeech-GL: Galician ASR corpus. Zenodo. PID https://doi.org/10.5281/zenodo.7… — [doi:10.5281/zenodo.7913218](https://doi.org/10.5281/zenodo.7913218) ↗
  15. Vladu, Adina Ioana and Marta Vázquez Abuín and Elisa Fernández Rei and Noelia García Díaz and Adrián Vidal Miguéns and Carmen Magariños. 2023. Nos_TranscriSpeech-GL: Galician ASR corpus. Zenodo. PID https://doi.org/10.5281/zenodo.7717140. Dataset. Available under CC BY 4.0 license. — [doi:10.5281/zenodo.7717140](https://doi.org/10.5281/zenodo.7717140) ↗
  16. Vladu, Adina Ioana and García Díaz, Noelia and Regueira Fernández, Xosé Luís and Magariños, Carmen and Moscoso Sánchez, Antonio and Fernández López, Daniel and Fernández Rei, Elisa and Dubert-García, Francisco. 2025. Nos_Brais-GL: Galician TTS corpus. Zenodo. PID https://doi.org/10.5281/zenodo.14265… — [doi:10.5281/zenodo.14265241](https://doi.org/10.5281/zenodo.14265241) ↗
  17. Vázquez Abuín, Marta and García Díaz, Noelia and Vladu, Adina Ioana and Magariños, Carmen and Vidal Miguéns, Adrián and Fernández Rei, Elisa. 2023. Nos_Celtia-GL: Galician TTS corpus (1.0.0.). Zenodo. find
  18. PID https://doi.org/10.5281/zenodo.7716958. Dataset. Available under CC BY 4.0 license. — [doi:10.5281/zenodo.7716958](https://doi.org/10.5281/zenodo.7716958) ↗

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/vladu-2026-building-collaborative-speech-corpora-low/