Elamur Research⌕
☰
home / papers / 2026 / Southern Kurdish Speech Recognition Resources and Benchmarking

Southern Kurdish Speech Recognition Resources and Benchmarking

Summary

Spoken by about 6 million people in western Iran and eastern Iraq, Southern Kurdish (ISO 639 sdh) is a Kurdish variety whose survival is at risk. It has almost no speech technology. It is not taught in schools, appears rarely in media, and differs enough from Central Kurdish that ASR systems built for Central Kurdish transfer poorly.

The authors build two resources to address this:

  • Bestun Corpus: a training set of roughly 30 hours of manually validated read speech. Native volunteers recorded sentences drawn from books and a social science magazine, after professional native editors had revised the text.
  • Evaluation benchmark: a multi-domain test set made by rewriting the 100-sentence Asosoft Central Kurdish test set into Southern Kurdish. Eight speakers of different vernaculars, none of whom contributed training data, recorded it.

The authors fine-tuned Whisper-turbo and wav2vec-bert (CTC) on the corpus. wav2vec-bert performed better, reaching 4.09 CER and 24.26 WER on the benchmark. An error analysis links the large gap between CER and WER mainly to inconsistent spelling, letters whose written form differs from their pronunciation, and variation between vernaculars. The work provides the first dedicated ASR data and baselines for this dialect.

Key points

  • Bestun Corpus: 18,636 validated recordings (15,273 unique sentences, about 1.55M tokens) from 208 speakers aged 18–49, stored as 24 kHz mono WAV. One speaker contributed 4,000 utterances. Recordings came from a Telegram bot (about 5 h) and a web tool.
  • Source text: about 400,000 tokens from 5 writers and 15 magazine volumes. About 16,000 sentences of 3–15 tokens were sampled and checked by four native editors, who removed Central Kurdish and Persian material.
  • Benchmark: 773 utterances (86 minutes) covering the Kermānshāhi, Kalhori, Malekshāhi, Garrusi and Kolyā'i vernaculars. Only two of the eight speakers are female.
  • Error analysis: per-speaker CER on identical sentences ranged from 3.02 to 6.49, and this spread did not correlate clearly with noise or reverberation. Errors fall into three types: standardization, phonetic and character mismatch (e.g., ح/ه, ع/ئ), and vernacular diversity (/ɫ/–/l/, /ɣ/–/q/).
  • Prior work and availability: the only earlier Southern Kurdish ASR dataset had 7 hours (Hameed et al., 2025). The new data and models are released under CC BY-NC-ND 4.0. Spontaneous speech is not yet covered.

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 10.63317/2rkqhw7hmo2d

@inproceedings{mohammadamini2026southern,
  title = {{Southern Kurdish Speech Recognition Resources and Benchmarking}},
  author = {Mohammadamini, Mohammad and Tahon, Marie},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/2rkqhw7hmo2d},
  url = {https://doi.org/10.63317/2rkqhw7hmo2d}
}

References

18 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Sina Ahmadi, Daban Jaff, Md Mahfuz Ibn Alam, and Antonios Anastasopoulos. 2024. Language and speech technology for Central Kurdish varieties. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 10034–1004… find
  2. 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),… find our card
  3. Sakina Azadi. 2022. A comparison between feyli and kalhori verbs of southern kurdish. Regional language and literature of Iran, 36(ISSUE):1–28. Accessed 9 October 2025. find
  4. Zahra Azin and Sina Ahmadi. 2021. Creating an electronic lexicon for the under-resourced southern varieties of kurdish language. In Proceedings of the Seventh Biennial Conference on Electronic Lexicography (eLex 2021), pages 479–488. find
  5. Loïc Barrault, Yu-An Chung, Mariano Cora Meglioli, David Dale, Ning Dong, Paul-Ambroise Duquenne, Hady Elsahar, Hongyu Gong, Kevin Heffernan, John Hoffman, Christopher Klaiber, Pengwei Li, Daniel Licht, Jean Maillard, Alice Rakotoarison, Kaushik Ram Sadagopan, Guillaume Wenzek, Ethan Ye, Bapi Akula,… find our card
  6. Loïc Barrault, Yu-An Chung, Mariano Coria Meglioli, David Dale, Ning Dong, Paul-Ambroise Duquenne, Hady Elsahar, Hongyu Gong, Kevin Heffernan, et. all., and SEAMLESS Communication Team. 2025. Joint speech and text machine translation for up to 100 languages. Nature, 637(8046):587–593. find
  7. Sara Belelli. 2019. Towards a dialectology of southern kurdish: Where to begin? In Songül Gündoğdu, Ergin Öpengin, Geoffrey Haig, and Erik Anonby, editors, Current issues in Kurdish linguistics, pages 73–92. University of Bamberg Press, Bamberg. find
  8. Sara Belelli. 2021. The Laki Variety of Harsin: Grammar, Texts, Lexicon. Number 2 in Bamberg Studies in Kurdish Linguistics. University of Bamberg Press, Bamberg. find
  9. 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. arXiv preprint arXiv:2108.06209. — [arXiv:2108.06209](https://arxiv.org/abs/2108.06209) find
  10. Ismaïl Kamandâr Fattah. 2000. Les dialectes kurdes méridionaux : étude linguistique et dialectologique. Number 37 in Acta Iranica. Peeters. find
  11. Alex Graves, Santiago Fernández, Faustino Gómez, and Jürgen Schmidhuber. 2006. Connectionist temporal classification: Labelling unsegmented sequence data with recurrent neural networks. In Proceedings of the 23rd International Conference on Machine Learning (ICML 2006). find
  12. 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
  13. Manijeh Mirmukri, Gholamhossein Karimi-Doostan, Yadgar Karimi, and Vahid Gholami. 2019. Mutual intelligibility between central and southern kurdish dialects (case study: Mahabadi and badrei varieties). Journal of Linguistics, Islamic Azad University, Sanandaj. Reçu le 24 janvier 2019 ; accepté le 24… find
  14. 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
  15. Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2022. Robust speech recognition via large-scale weak supervision. arXiv preprint arXiv:2212.04356. find our card
  16. Jaffer Sheyholislami. 2008. Identity, Discourse, and the Media: The Case of the Kurds. Phd dissertation, Carleton University. Accessed 9 October 2025. find
  17. Jaffer Sheyholislami. 2015. The Kurds: History, Religion, Language, Politics, chapter Language Varieties of the Kurds. Austrian Federal Ministry of the Interior. find
  18. Hadi Veisi, Hawre Hosseini, Mohammad MohammadAmini, Wirya Fathy, and Aso Mahmudi. 2022. Jira: a central kurdish speech recognition system, designing and building speech corpus and pronunciation lexicon. Language Resources and Evaluation, 56(3):917–941. find

↑ back to top