---
title: "IQRA 2026: Interspeech Challenge on Automatic — Kheir 2026"
description: "Arabic mispronunciation detection and diagnosis (MDD) has long lacked open data and comparable evaluation, unlike English or Mandarin."
authors: ["Yassine El Kheir", "Amit Meghanani", "Mostafa Shahin", "Omnia Ibrahim", "Shammur Absar Chowdhury", "Nada AlMarwani", "Youssef Elshahawy", "Ahmed Ali"]
year: 2026
venue: "arXiv:2603.29087"
source: "https://arxiv.org/abs/2603.29087"
doi: "10.48550/arxiv.2603.29087"
arxiv: "2603.29087"
licence: "CC BY 4.0 (arXiv)"
licence_url: "http://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/kheir-2026-iqra-2026-interspeech-challenge-automatic/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/kheir-2026-iqra-2026-interspeech-challenge-automatic/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/6/) / IQRA 2026: Interspeech Challenge on Automatic Pronunciation Assessment for Modern Standard Arabic (MSA)

# IQRA 2026: Interspeech Challenge on Automatic Pronunciation Assessment for Modern Standard Arabic (MSA)

**Authors:** [Yassine El Kheir](https://md.research.elamur.ai/free/en/authors/yassine-el-kheir/), [Amit Meghanani](https://md.research.elamur.ai/free/en/authors/amit-meghanani/), [Mostafa Shahin](https://md.research.elamur.ai/free/en/authors/mostafa-shahin/), [Omnia Ibrahim](https://md.research.elamur.ai/free/en/authors/omnia-ibrahim/), [Shammur Absar Chowdhury](https://md.research.elamur.ai/free/en/authors/shammur-absar-chowdhury/), [Nada AlMarwani](https://md.research.elamur.ai/free/en/authors/nada-almarwani/), Youssef Elshahawy, [Ahmed Ali](https://md.research.elamur.ai/free/en/authors/ahmed-ali/) · **Year:** 2026 · **Venue:** arXiv:2603.29087 · **Licence:** [CC BY 4.0 (arXiv)](http://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2603.29087) · [PDF](https://arxiv.org/pdf/2603.29087) · [DOI 10.48550/arxiv.2603.29087](https://doi.org/10.48550/arxiv.2603.29087) · [arXiv 2603.29087](https://arxiv.org/abs/2603.29087) · [OpenAlex](https://openalex.org/W7147583859)

**Citations:** 0 citations (openalex-snapshot · 2026-10-14)

## Summary
Arabic mispronunciation detection and diagnosis (MDD) has long lacked open data and comparable evaluation, unlike English or Mandarin. This paper reports the second IQRA shared challenge for Modern Standard Arabic, held at Interspeech 2026. Each system gets an utterance and its vowelized reference text and must output the phonemes the speaker actually produced. Outputs use a 68-phoneme inventory from the Halabi phonetizer and are scored against both the canonical sequence and the annotated verbatim one. The organizers provided three training resources: the open Iqra train corpus, the synthetic Iqra TTS set with injected errors, and the new Iqra Extra IS26 set of genuine human mispronunciations. Systems were tested on the expanded QuranMB.v2 benchmark. Nineteen teams took part, using CTC models on self-supervised encoders, staged fine-tuning, language-model rescoring and large audio-language models. The winning F1 was 0.7201, about 0.28 above the first edition's best and well above the mHuBERT baseline (0.4414). The challenge gives the field a shared, reproducible yardstick, so progress in Arabic pronunciation assessment can now be measured.

## Key points
- Training data:
  - Iqra train: about 79h and 74k utterances, built from Common Voice Arabic v12.0 plus Qur'anic recitation, with expert vowelization.
  - Iqra TTS: about 52h and 55.4k utterances from seven single-speaker TTS voices. Half are canonical and half contain errors drawn from a phoneme confusion matrix.
  - Iqra Extra IS26: 1,333 utterances (about 1.5h) of real speakers deliberately producing specified phoneme substitutions.
- Test set: QuranMB.v2, with 1,643 utterances (about 2.5h) whose phoneme annotations were checked by three Arabic linguistic experts.
- Metrics: F1 over correctly detected mispronunciations is the ranking metric, with PER reported alongside. Detected errors are further split into correct and erroneous diagnosis.
- Baseline: frozen mHuBERT (147 languages, 94M parameters) with a SUPERB-style weighted layer sum, a 2-layer Bi-LSTM and CTC loss, scoring F1 0.4414.
- Top systems:
  - whu-iasp (0.7201): wav2vec2-xls-r-300m encoder, a TCN and a two-stage curriculum.
  - UTokyo (0.7170): CROTTC, which uses optimal-transport alignment and consistency regularization.
  - RAM (0.7157): focused on label reliability rather than a new architecture.

## Interesting ideas
- The winning team first learned a general mapping on large real plus synthetic data, then adapted on the small set of authentic error data, so scarce genuine examples are spent where they matter most. — For Mova (learn): train parsers on abundant silver data first, then run a short final pass on the small hand-checked gold portion rather than mixing them uniformly.
- Synthetic errors were not random but sampled from a confusion matrix of how people actually tend to err, which makes artificial negatives look realistic. — For Mova (learn): build negative controls by corrupting tags or arcs according to the parser's observed confusion statistics, so tests reflect plausible failures rather than trivial noise.
- Scoring separates two questions: whether an error was spotted, and whether the right correction was named. A system can flag a problem yet misdiagnose it. — For Mova (learn): when evaluating self-induced repair rules, report separately whether a rule fired on a truly bad structure and whether its fix matched gold.
- Outputs from several checkpoints were merged with weighted edit-distance voting before a language model rescored them, turning disagreement into a more reliable consensus. — For Mova (tree): combine perceptron and gparse outputs through weighted arc-level voting, keeping each witness's vote visible for explanation.
- Consistency regularization perturbs the input and pushes the model to give matching predictions, which makes it less sensitive to noise. — For Mova (tree): perturb sentences harmlessly (casing, punctuation, synonym swaps) and penalize or flag cases where the UD tree changes beyond the edited tokens.

*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](https://md.research.elamur.ai/free/en/shelves/linguistics/)

## Cite

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

```bibtex
@misc{kheir2026iqra,
  title = {{IQRA 2026: Interspeech Challenge on Automatic Pronunciation Assessment for Modern Standard Arabic (MSA)}},
  author = {Kheir, Yassine El and Meghanani, Amit and Shahin, Mostafa and Ibrahim, Omnia and Chowdhury, Shammur Absar and AlMarwani, Nada and Elshahawy, Youssef and Ali, Ahmed},
  year = {2026},
  doi = {10.48550/arxiv.2603.29087},
  eprint = {2603.29087},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2603.29087}
}
```

- [← previous Investigating Reasoning with Hypotheses: The RIP2 Corpus](https://md.research.elamur.ai/free/en/papers/schad-2026-investigating-reasoning-hypotheses-rip2-corpus/)
- [next → Is AI Catching Up to Human Expression? Exploring Emotion, Personality, Authorship, and Linguistic Style in English and Arabic with Six Large Language Models](https://md.research.elamur.ai/free/en/papers/alsadhan-2026-ai-catching-up-human-expression/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/6/)

## References

19 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. A. Neri, O. Mich, M. Gerosa, and D. Giuliani, “The effectiveness of computer assisted pronunciation training for foreign language learning by children,” Computer Assisted Language Learning, vol. 21, no. 5, pp. 393–408, 2008. [find](https://scholar.google.com/scholar?q=A%2E+Neri%2C+O%2E+Mich%2C+M%2E+Gerosa%2C+and+D%2E+Giuliani%2C+%E2%80%9CThe+effectiveness+of+computer+assisted+pronunciation+training+for+foreign+language+learning+by+children%2C%E2%80%9D+Computer+Assisted+Language+Learning%2C+vol%2E+21%2C+n)
2. P. M. Rogerson-Revell, “Computer-assisted pronunciation training (capt): Current issues and future directions,” Relc Journal, vol. 52, no. 1, pp. 189–205, 2021. [find](https://scholar.google.com/scholar?q=P%2E+M%2E+Rogerson%2DRevell%2C+%E2%80%9CComputer%2Dassisted+pronunciation+training+%28capt%29%3A+Current+issues+and+future+directions%2C%E2%80%9D+Relc+Journal%2C+vol%2E+52%2C+no%2E+1%2C+pp%2E+189%E2%80%93205%2C+2021%2E)
3. Y. E. Kheir, A. Ali, and S. A. Chowdhury, “Automatic pronunciation assessment–a review,” arXiv preprint arXiv:2310.13974, 2023. [↗](https://arxiv.org/abs/2310.13974)
4. K. Li, X. Qian, and H. Meng, “Mispronunciation detection and diagnosis in l2 english speech using multidistribution deep neural networks,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 25, no. 1, pp. 193–207, 2016. [find](https://scholar.google.com/scholar?q=K%2E+Li%2C+X%2E+Qian%2C+and+H%2E+Meng%2C+%E2%80%9CMispronunciation+detection+and+diagnosis+in+l2+english+speech+using+multidistribution+deep+neural+networks%2C%E2%80%9D+IEEE%2FACM+Transactions+on+Audio%2C+Speech%2C+and+Language+Processi)
5. W.-K. Leung, X. Liu, and H. Meng, “Cnn-rnn-ctc based end-to-end mispronunciation detection and diagnosis,” in ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2019, pp. 8132–8136. [find](https://scholar.google.com/scholar?q=W%2E%2DK%2E+Leung%2C+X%2E+Liu%2C+and+H%2E+Meng%2C+%E2%80%9CCnn%2Drnn%2Dctc+based+end%2Dto%2Dend+mispronunciation+detection+and+diagnosis%2C%E2%80%9D+in+ICASSP+2019%2D2019+IEEE+International+Conference+on+Acoustics%2C+Speech+and+Signal+Processing+)
6. X. Qian, H. Meng, and F. Soong, “Capturing L2 segmental mispronunciations with joint-sequence models in CAPT,” in Proc. Interspeech, 2010. [find](https://scholar.google.com/scholar?q=X%2E+Qian%2C+H%2E+Meng%2C+and+F%2E+Soong%2C+%E2%80%9CCapturing+L2+segmental+mispronunciations+with+joint%2Dsequence+models+in+CAPT%2C%E2%80%9D+in+Proc%2E+Interspeech%2C+2010%2E)
7. S. Fort et al., “Evaluating Wav2Vec2-BERT for CAPT in low-resource languages,” in Proc. Interspeech, 2025. [find](https://scholar.google.com/scholar?q=S%2E+Fort+et+al%2E%2C+%E2%80%9CEvaluating+Wav2Vec2%2DBERT+for+CAPT+in+low%2Dresource+languages%2C%E2%80%9D+in+Proc%2E+Interspeech%2C+2025%2E)
8. Y. E. Kheir, F. Khnaisser, S. A. Chowdhury, H. Mubarak, S. Afzal, and A. Ali, “Qvoice: Arabic speech pronunciation learning application,” arXiv preprint arXiv:2305.07445, 2023. [20] Y. El Kheir, A. Meghanani, H. O. Toyin, N. Almarwani, O. Ibrahim, Y. A. Elshahawy, M. Shahin, and A. Ali, “Iqra’eval: A shared task on qur’anic pronunciation assessment,” in Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, 2025, pp. 443–452. [21] Y. El Kheir, O. Ibrahim, A. Meghanani, N. Almarwani, H. Toyin, S. Alharbi, M. Alfadly, L. Alkanhal, I. Selim, S. Elbatal, S. Mdhaffar, T. Hain, Y. Hifny, M. Shahin, and A. Ali, “Towards a Unified Benchmark for Arabic Pronunciation Assessment: Qur’anic Recitation as Case Study,” in Interspeech 2025, 2025, pp. 2410– 2414. [22] N. Halabi and M. Wald, “Phonetic inventory for an arabic speech corpus,” in Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC’16), 2016, pp. 734– 738. [23] R. Ardila, M. Branson, K. Davis, M. Henretty, M. Kohler, J. Meyer, R. Morais, L. Saunders, F. M. Tyers, and G. Weber, “Common voice: A massively-multilingual speech corpus,” arXiv preprint arXiv:1912.06670, 2019. [24] M. Zanon Boito, V. Iyer, N. Lagos, L. Besacier, and I. Calapodescu, “mhubert-147: A compact multilingual hubert model,” in Interspeech 2024, 2024, pp. 3939–3943. [25] S. wen Yang, P.-H. Chi, Y.-S. Chuang, C.-I. J. Lai, and K. L. et al., “Superb: Speech processing universal performance benchmark,” in Interspeech 2021, 2021, pp. 1194–1198. [↗](https://arxiv.org/abs/2305.07445) [our card](https://md.research.elamur.ai/free/en/papers/kheir-2023-qvoice-arabic-speech-pronunciation-learning/)
9. Y. E. Kheir, H. Mubarak, A. Ali, and S. A. Chowdhury, “Beyond orthography: Automatic recovery of short vowels and dialectal sounds in arabic,” arXiv preprint arXiv:2408.02430, 2024. [26] A. Graves, S. Fernández, F. Gomez, and J. Schmidhuber, “Connectionist temporal classification: labelling unsegmented sequence data with recurrent neural networks,” in ICML, ser. ICML ’06. NY, USA: Association for Computing Machinery, 2006, p. 369–376. [Online]. Available: https://doi.org/10.1145/1143844.1143891 [↗](https://doi.org/10.1145/1143844.1143891) [our card](https://md.research.elamur.ai/free/en/papers/acl-2024-kheir-beyond-orthography-automatic-recovery-short/)
10. N. Alrashoudi, H. Al-Khalifa, and Y. Alotaibi, “Improving mispronunciation detection and diagnosis for non-native learners of the arabic language,” Discover Computing, vol. 28, no. 1, p. 1, 2025. [27] C. Lea, M. D. Flynn, R. Vidal, A. Reiter, and G. D. Hager, “Temporal convolutional networks for action segmentation and detection,” in proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 156–165. [find](https://scholar.google.com/scholar?q=N%2E+Alrashoudi%2C+H%2E+Al%2DKhalifa%2C+and+Y%2E+Alotaibi%2C+%E2%80%9CImproving+mispronunciation+detection+and+diagnosis+for+non%2Dnative+learners+of+the+arabic+language%2C%E2%80%9D+Discover+Computing%2C+vol%2E+28%2C+no%2E+1%2C+p%2E+1%2C+2025%2E+%5B27%5D)
11. J. Zhang, C. Ni, S. Zhang et al., “Speechocean762: An open-source non-native English speech corpus for pronunciation assessment,” in Proc. Interspeech, 2021. [28] F. James, “Modified kneser-ney smoothing of n-gram models,” Research Institute for Advanced Computer Science, Tech. Rep. 00.07, 2000. [find](https://scholar.google.com/scholar?q=J%2E+Zhang%2C+C%2E+Ni%2C+S%2E+Zhang+et+al%2E%2C+%E2%80%9CSpeechocean762%3A+An+open%2Dsource+non%2Dnative+English+speech+corpus+for+pronunciation+assessment%2C%E2%80%9D+in+Proc%2E+Interspeech%2C+2021%2E+%5B28%5D+F%2E+James%2C+%E2%80%9CModified+kneser%2Dney+smooth)
12. D. Y. Zhang, S. Saha, and S. Campbell, “Phonetic RNN-Transducer for mispronunciation detection,” in Proc. ICASSP, 2023. [29] A. Baevski, Y. Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representations,” in Advances in Neural Information Processing Systems, H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin, Eds., vol. 33. Curran Associates, Inc., 2020, pp. 12 449–12 460. [find](https://scholar.google.com/scholar?q=D%2E+Y%2E+Zhang%2C+S%2E+Saha%2C+and+S%2E+Campbell%2C+%E2%80%9CPhonetic+RNN%2DTransducer+for+mispronunciation+detection%2C%E2%80%9D+in+Proc%2E+ICASSP%2C+2023%2E+%5B29%5D+A%2E+Baevski%2C+Y%2E+Zhou%2C+A%2E+Mohamed%2C+and+M%2E+Auli%2C+%E2%80%9Cwav2vec+2%2E0%3A+A+framework+for) [our card](https://md.research.elamur.ai/free/en/papers/baevski-2020-wav2vec-2-0-framework-self/)
13. Ş. S. Çalık, A. Küçükmanisa, and Z. H. Kilimci, “A novel framework for mispronunciation detection of arabic phonemes using audio-oriented transformer models,” Applied Acoustics, vol. 215, p. 109711, 2024. [find](https://scholar.google.com/scholar?q=S%CC%A7%2E+S%2E+C%CC%A7al%C4%B1k%2C+A%2E+Ku%CC%88c%CC%A7u%CC%88kmanisa%2C+and+Z%2E+H%2E+Kilimci%2C+%E2%80%9CA+novel+framework+for+mispronunciation+detection+of+arabic+phonemes+using+audio%2Doriented+transformer+models%2C%E2%80%9D+Applied+Acoustics%2C+vol%2E+215%2C+p%2E+1097)
14. A. Baevski, Y. Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representations,” Advances in neural information processing systems, vol. 33, pp. 12 449–12 460, 2020. [find](https://scholar.google.com/scholar?q=A%2E+Baevski%2C+Y%2E+Zhou%2C+A%2E+Mohamed%2C+and+M%2E+Auli%2C+%E2%80%9Cwav2vec+2%2E0%3A+A+framework+for+self%2Dsupervised+learning+of+speech+representations%2C%E2%80%9D+Advances+in+neural+information+processing+systems%2C+vol%2E+33%2C+pp%2E+12+449%E2%80%93) [our card](https://md.research.elamur.ai/free/en/papers/baevski-2020-wav2vec-2-0-framework-self/)
15. W. Hsu, B. Bolte, Y. Tsai, K. Lakhotia, R. Salakhutdinov, and A. Mohamed, “Hubert: Self-supervised speech representation learning by masked prediction of hidden units,” IEEE/ACM Trans. Audio, Speech and Lang. Proc., vol. 29, p. 3451–3460, oct 2021. [Online]. Available: https://doi.org/10.1109/TASLP.2021. 3122291 [↗](https://doi.org/10.1109/taslp.2021) [our card](https://md.research.elamur.ai/free/en/papers/hsu-2021-hubert-self-supervised-speech-representation/)
16. S. Kim et al., “Automatic pronunciation assessment using self-supervised speech representation learning,” in Proc. Interspeech, 2022. [find](https://scholar.google.com/scholar?q=S%2E+Kim+et+al%2E%2C+%E2%80%9CAutomatic+pronunciation+assessment+using+self%2Dsupervised+speech+representation+learning%2C%E2%80%9D+in+Proc%2E+Interspeech%2C+2022%2E)
17. R. C. Shekar, M. Yang, K. Hirschi et al., “Assessment of non-native speech intelligibility using Wav2vec2-based MDD and multi-level GOP transformer,” in Proc. Interspeech, 2023. [find](https://scholar.google.com/scholar?q=R%2E+C%2E+Shekar%2C+M%2E+Yang%2C+K%2E+Hirschi+et+al%2E%2C+%E2%80%9CAssessment+of+non%2Dnative+speech+intelligibility+using+Wav2vec2%2Dbased+MDD+and+multi%2Dlevel+GOP+transformer%2C%E2%80%9D+in+Proc%2E+Interspeech%2C+2023%2E)
18. A. Ali, P. Bell, J. Glass, Y. Messaoui, H. Mubarak, S. Renals, and Y. Zhang, “The mgb-2 challenge: Arabic multi-dialect broadcast media recognition,” in 2016 IEEE Spoken Language Technology Workshop (SLT). IEEE, 2016, pp. 279–284. [find](https://scholar.google.com/scholar?q=A%2E+Ali%2C+P%2E+Bell%2C+J%2E+Glass%2C+Y%2E+Messaoui%2C+H%2E+Mubarak%2C+S%2E+Renals%2C+and+Y%2E+Zhang%2C+%E2%80%9CThe+mgb%2D2+challenge%3A+Arabic+multi%2Ddialect+broadcast+media+recognition%2C%E2%80%9D+in+2016+IEEE+Spoken+Language+Technology+Workshop+%28) [our card](https://md.research.elamur.ai/free/en/papers/ali-2016-mgb-2-challenge-arabic-multi/)
19. P. Gerber et al., “Non-native children’s automatic speech assessment challenge (NOCASA),” in Proc. IEEE MLSP, 2025. [30] D. Wu, B. Zhang, C. Yang, Z. Peng, W. Xia, X. Chen, and X. Lei, “U2++: Unified two-pass bidirectional end-to-end model for speech recognition,” arXiv preprint arXiv:2106.05642, 2021. [31] J. Kim, J. Kong, and J. Son, “Conditional variational autoencoder with adversarial learning for end-to-end text-to-speech,” 2021. [Online]. Available: https://arxiv.org/abs/2106.06103 [32] D. S. Park, W. Chan, Y. Zhang, C.-C. Chiu, B. Zoph, E. D. Cubuk, and Q. V. Le, “Specaugment: A simple data augmentation method for automatic speech recognition,” arXiv preprint arXiv:1904.08779, 2019. [33] A. Conneau, M. Ma, S. Khanuja, Y. Zhang, V. Axelrod, S. Dalmia, J. Riesa, C. Rivera, and A. Bapna, “Fleurs: Few-shot learning evaluation of universal representations of speech,” in 2022 IEEE Spoken Language Technology Workshop (SLT). IEEE, 2023, pp. 798–805. [34] R. Ardila, M. Branson, K. Davis, M. Kohler, J. Meyer, M. Henretty, R. Morais, L. Saunders, F. Tyers, and G. Weber, “Common voice: A massively-multilingual speech corpus,” in Proceedings of the twelfth language resources and evaluation conference, 2020, pp. 4218–4222. [↗](https://arxiv.org/abs/2106.05642) [our card](https://md.research.elamur.ai/free/en/papers/park-2019-specaugment-simple-data-augmentation-method/)