Elamur Research⌕
☰
home / papers / 2026 / Exploring the reusability of Northern Kurdish resources for Badini speech recognition

Exploring the reusability of Northern Kurdish resources for Badini speech recognition

Paper · 2026 · Mohammad Mohammadamini, Aveen Jalal Mohammed, Barzan Hussein Mohammed et al.

Summary

Badini is the Northern Kurdish (KMR) variant of Iraq's Duhok province. It is written in a modified Arabic script, like Central Kurdish (CKB), and before this work it had no dedicated ASR resources. The authors test how far Latin-script KMR resources can be adapted for it.

They converted the KMR portion of Common Voice 18 into Arabic script with the AsoSoft library. They then corrected word joining and splitting by hand to follow Arabic-script Kurdish orthography. They also crawled a new Badini text collection and turned around 119k of its sentences into synthetic speech with the MMS Northern Kurdish TTS model. For evaluation, they recorded a multi-domain read-speech benchmark.

They fine-tuned Whisper V3 Large and Wav2Vec2.0-BERT (trained with CTC). Training on transliterated and synthetic speech together worked best, reaching 6.80% CER and 34.41% WER. Error analysis traces the remaining gap to two sources:

  • letters that are lost during transliteration;
  • lexical and morphosyntactic differences between Badini and standard KMR.

The study shows that adapting existing data helps but is not enough, and that Badini needs resources of its own.

Key points

  • Three resources, all released under CC BY-NC-ND 4.0 on Hugging Face (BadiniSpeechNLP):
    • transliterated Common Voice KMR: 16,808 unique sentences, 61.6k utterances, 68h;
    • Badini text corpus: 14.22M tokens from 70k documents across 15 websites;
    • ASR benchmark: 623 sentences, 1,028 recordings by 5 speakers (2 male, 3 female), 2h46m, eight topics.
  • Synthetic data: sentences of 3–20 tokens were synthesized with MMS TTS, yielding 207h of speech.
  • Results by training data:
    • Whisper trained on CV-CKB performed poorly (28.22 CER / 76.65 WER).
    • Wav2Vec-BERT reached 11.33 CER on CV-KMR alone and 9.25 CER on synthetic Badini speech alone.
    • Combining both gave 6.80 CER, a 39% relative CER reduction over real KMR data alone.
  • Per-domain results (Wav2Vec-BERT): politics was easiest (5.29 CER), likely because much of the synthetic text comes from that domain. The science domain was harder, which the authors link to its many non-Kurdish terms.
  • Lost letters: the Arabic-script letters ح, ع, غ, ڵ have no Latin-script equivalents. The best model predicts them two to three times less often than they occur in the reference transcripts.

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

🎓Le Mans UniversitéLe Mans · France15 works · 31,097 citations🎓University of DuhokDuhok · Iraq2 works · 1 citation

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSignal Processing · 452 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 2,908 worksSpeech & spoken language · 480 works

Cite

Cite as Mohammad Mohammadamini, Aveen Jalal Mohammed, Barzan Hussein Mohammed, Dezheen H. Abdulazeez, Imad Saeed Sadeeq, Dilgash Mohammed Salih, Amera Ismail Melhum and Abuobaida Abdullah Dheyab. 2026. Exploring the reusability of Northern Kurdish resources for Badini speech recognition. Proceedings of the First Workshop on Dialects in NLP –- A Resource Perspective.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/2gzjngtqiqp7

@inproceedings{mohammadamini2026exploring,
  title = {{Exploring the reusability of Northern Kurdish resources for Badini speech recognition}},
  author = {Mohammad Mohammadamini, Aveen Jalal Mohammed, Barzan Hussein Mohammed, Dezheen H. Abdulazeez, Imad Saeed Sadeeq, Dilgash Mohammed Salih, Amera Ismail Melhum and Abuobaida Abdullah Dheyab},
  booktitle = {Proceedings of the First Workshop on Dialects in NLP –- A Resource Perspective},
  year = {2026},
  doi = {10.63317/2gzjngtqiqp7},
  url = {https://doi.org/10.63317/2gzjngtqiqp7}
}

References

18 references, 5 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, Mike Kohler, Josh Meyer, Reuben Morais, Lindsay Saunders, Francis M. Tyers, and Gregor Weber. 2020. Common voice: A massively-multilingual speech corpus. In Proceedings of the 12th Language Resources and Evaluation Conference (LREC 2020),…
  2. Verena Blaschke, Miriam Winkler, Constantin Förster, Gabriele Wenger-Glemser, and Barbara Plank. 2025. A Multi-Dialectal Dataset for German Dialect ASR and Dialect-to-Standard Speech Translation. In Interspeech 2025, pages 913–917.
  3. Yu-An Chung, Yu Zhang, Wei Han, Chung-Cheng Chiu, James Qin, Ruoming Pang, and Yonghui Wu. 2021. W2v-BERT: Combining contrastive learning and masked language modeling for self-supervised speech pre-training. In Proceedings of the 2021 IEEE Automatic Speech Recognition and Understanding Workshop (ASR… find
  4. Alexis Conneau, Min Ma, Simran Khanuja, Yu Zhang, Vera Axelrod, Siddharth Dalmia, Jason Riesa, Clara Rivera, and Ankur Bapna. 2023. Fleurs: Few-shot learning evaluation of universal representations of speech. In 2022 IEEE Spoken Language Technology Workshop (SLT), pages 798–805.
  5. Geoffrey Haig. 2018. 2.3. northern kurdish (kur-manjî). The languages and linguistics of Western Asia: An areal perspective, 6:106. find
  6. Razhan Hameed, Sina Ahmadi, Hanah Hadi, and Rico Sennrich. 2025. Automatic Speech Recognition for Low-Resourced Middle Eastern Languages. In Interspeech 2025, pages 733–737. find
  7. A Mahmudi, H Veisi, M MohammadAmini, and H Hosseini. 2019. Automated kurdish text normalization. In The second international conference on kurdish and persian languages and literature. find
  8. Aso Mahmudi and Hadi Veisi. 2021. Automated grapheme-to-phoneme conversion for central kurdish based on optimality theory. Computer Speech Language, 70:101222. find
  9. Yaron Matras. 2019. Revisiting Kurdish dialect geography: Findings from the Manchester database. In Current Issues in Kurdish Linguistics, pages 225–241. University of Bamberg Press, Bamberg. find
  10. Mohammad Mohammadamini, Aghilas Sini, Marie Tahon, and Antoine Laurent. 2025. Scaling pseudo-labeling data for end-to-end low-resource speech translation (the case of Kurdish language). In Interspeech 2025, pages 898–902. find
  11. Ergin Öpengin and Geoffrey Haig. 2014. Regional variation in kurmanji: A preliminary classification of dialects. In Kurdish Studies Archive, pages 171–212. Brill. find
  12. Vineel Pratap, Andros Tjandra, Bowen Shi, Paden Tomasello, Arun Babu, Sayani Kundu, Ali Elkahky, Zhaoheng Ni, Apoorv Vyas, Maryam Fazel-Zarandi, Alexei Baevski, Yossi Adi, Xiaohui Zhang, Wei-Ning Hsu, Alexis Conneau, and Michael Auli. 2024. Scaling speech technology to 1,000+ languages. Journal of M…
  13. Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2022. Whisper: Robust speech recognition via large-scale weak supervision. arXiv preprint arXiv:2212.04356.
  14. Jaffer Sheyholislami. 2015. The Kurds: History, Religion, Language, Politics, chapter Language Varieties of the Kurds. Austrian Federal Ministry of the Interior. find
  15. Badini Orthography Team. 2025. Renusi Badini (In Kurdish). University of Duhok. find
  16. Terry Lynn Todd. 1985. A Grammar of Dimili, also known as Zaza. Ph.d. dissertation, University of Michigan, Ann Arbor, MI. Near Eastern Studies: Languages and Literatures. find
  17. H. Veisi, H. Hosseini, M. MohammadAmini, et al. 2022. Jira: a central kurdish speech recognition system, designing and building speech corpus and pronunciation lexicon. Language Resources and Evaluation, 56:917–941. find
  18. Hadi Veisi, Mohammad MohammadAmini, and Hawre Hosseini. 2020. Toward kurdish language processing: Experiments in collecting and processing the asosoft text corpus. Digital Scholarship in the Humanities, 35(1):176–193. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mohammadamini-2026-exploring-reusability-northern-kurdish-resources/