Elamur Research⌕
☰
home / papers / 2020 / Smartajweed Automatic Recognition of Arabic Quranic Recitation Rules

Smartajweed Automatic Recognition of Arabic Quranic Recitation Rules

Summary

The paper addresses automatic checking of Tajweed, the rules governing how Quranic Arabic is pronounced during recitation, such as merging, concealing, lightening or thinning certain letters. The authors argue that earlier systems mostly verified basic pronunciation or worked only on specific surahs. Instead, their system can check any verse against four particular rules: the two Meem rules, Edgham and Ekhfaa, and the two Lam rules, takhfeef and Tarqeeq. Expert speakers recorded each rule both correctly and in typical incorrect forms, from native and non-native Arabic speakers. The audio is resampled, normalised to a fixed length, and turned into filter-bank features. A separate SVM for each rule decides between right and wrong pronunciation. At inference time, a sliding window passes over the user's recording, and per-class thresholds filter which predictions count. The authors report about 99% accuracy. The system runs in the beta version of the IMAM learning app, which also gathers new user recordings for expert labelling. The work matters as a practical self-study tool for learners who lack reliable access to a teacher.

Key points

  • Four Tajweed rules, each with a correct and an incorrect variant, give 8 classes. The initial dataset has 657 recordings, roughly 80 per class, from universities, volunteer experts and paid experts.
  • The beta app gathered about 600 additional user recordings in two months, and an expert reviews and labels every one.
  • Preprocessing:
    • Clips are trimmed by hand to the rule segment and resampled to 8000 Hz.
    • Each clip is padded with silence or randomly truncated to 4 seconds, the average clip length.
  • Features:
    • Frames are 25 ms long with a 10 ms stride and a Hamming window.
    • An FFT magnitude spectrum is passed through 70 triangular band-pass filters.
  • Inference:
    • The rule name selects that rule's SVM model.
    • The audio is scanned in 4 s windows with a 0.5 s stride.
    • Each prediction is kept only if it exceeds a class-specific threshold.
    • The reported accuracy is about 99%.

Interesting ideas

  • I1. The caller names which rule to check, and a dedicated small model is loaded for that rule alone, so each decision stays narrow and auditable. — For Mova (answer): A reader could route each question type, such as a counting problem or a who-did-what question, to a dedicated checker, so every verdict traces back to one named specialist.
  • I2. Each class gets its own confidence threshold, and predictions below it are discarded rather than forced into a label. This gives tunable control over precision and the option to abstain. — For Mova (frames): Per-role thresholds on frame or role assignments would let prag leave a role unfilled when evidence is weak, and the explanation could show which threshold was missed.
  • I3. Training data deliberately includes the plausible wrong ways to perform each rule, not just correct examples, so the classifier learns the specific confusions that matter. — For Mova (learn): Curated "typical wrong" variants, such as near-miss parses or misread attachments, could serve as negative controls in evaluation, beside the gold examples.
  • I4. A fixed-size window slides over unsegmented input, and the best-supported window is chosen, which locates the relevant span without a separate segmenter. — For Mova (tokens): A similar span scan with per-pattern scores could find multiword expressions or idioms in raw token streams before lemmatisation.
  • I5. Deployment doubles as data collection: every user input and prediction is stored, reviewed by an expert, and fed into periodic retraining. — For Mova (learn): Logging real queries together with Mova's explained outputs, then having a teacher tier label the disagreements, would create a steady stream of hard silver and gold cases for repair rules.

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

Topics: Semitic languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2101.04200 · arXiv 2101.04200

@misc{alagrami2020smartajweed,
  title = {{Smartajweed Automatic Recognition of Arabic Quranic Recitation Rules}},
  author = {Alagrami, Ali M. and Eljazzar, Maged M.},
  year = {2020},
  doi = {10.48550/arxiv.2101.04200},
  eprint = {2101.04200},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2101.04200}
}

References

24 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Y. O. M. Elhadj, M. Aoun-Allah, I. A. Alsughaiyer, and A. Alansari, "A New Scientific Formulation of Tajweed Rules for E-Learning of Quran Phonological Rules," Edited by Anderson Silva, Elvis Pontes, p. 197, 2012. find
  2. A. Alfaries, M. Albahlal, M. Almazrua, and A. Almazrua, "A rule-based annotation system to extract tajweed rules from quran," in 2013 Taibah University International Conference on Advances in Information Technology for the Holy Quran and Its Sciences, 2013: IEEE, pp. 281- 286. find
  3. I. Ahsiah, N. Noor, and M. Idris, "Tajweed checking system to support recitation," in 2013 International Conference on Advanced Computer Science and Information Systems (ICACSIS), 2013: Ieee, pp. 189-193. find
  4. N. J. Ibrahim, M. Y. I. Idris, M. Z. M. Yusoff, and A. Anuar, "The Problems, Issues and Future Challenges of Automatic Speech Recognition for Quranic verse Recitation: A Review," Al-Bayan: Journal of Qur'an and Hadith Studies, vol. 13, no. 2, pp. 168-196, 2015. find
  5. S. S. B. Hassan and M. A. B. Zailaini, "Analysis of Tajweed errors in Quranic recitation," in Procedia-Social and Behavioral Sciences, 2013, vol. 103, no. Tq 1000: Elsevier, pp. 136-145. find
  6. A. Awang, A. Che Mat, and A. N. Yaakub, "Kebolehupayaan membaca al-Quran di kalangan pelajar diploma UiTM Terengganu: satu kajian kes," 2010. find
  7. A. R. Ali Hasan, S. Sabarudin, and C. B. Che Mat, "Tahap penguasaan bacaan Al-Quran dikalangan pelajar-pelajar UiTM Sarawak," 2010. find
  8. N. Kamarudin and S. Salam, "Tajweed understanding among Malaysian muslim (a preliminary findings)," in Proc. 1st International Conference on Mobile Learning, Applications and Services (Mobilecase2012), Kuala Lumpur, 2012, 2012, pp. 1-4. find
  9. A. H. Tamuri, A. M. I. Amal, H. M. Noor, and M. I. M. Pisol, "Penilaian Guru terhadap Pelaksanaan Model-Model Tilawah al-Quran Program j-QAF/Teachers Evaluation on the Implementation of j-QAF Quranic Recitation Models," International Journal of Islamic Thought, vol. 3, p. 1, 2013. find
  10. A. N. Akkila and S. S. Abu-Naser, "Rules of Tajweed the Holy Quran Intelligent Tutoring System," 2018. find
  11. S. M. Abdou and M. Rashwan, "A Computer Aided Pronunciation Learning system for teaching the holy quran Recitation rules," in 2014 IEEE/ACS 11th International Conference on Computer Systems and Applications (AICCSA), 2014: IEEE, pp. 543-550. find
  12. A. A. Y. AlZoubi, "Use of Information Technology in the Teaching of Quran Recitation (Qira'at)-Electronic Miqrah as a Model," in 2013 Taibah University international conference on advances in information technology for the Holy Quran and its sciences, 2013: IEEE, pp. 593-621. find
  13. W. M. Muhammad, R. Muhammad, A. Muhammad, and A. Martinez-Enriquez, "Voice content matching system for quran readers," in 2010 Ninth Mexican International Conference on Artificial Intelligence, 2010: IEEE, pp. 148-153. find
  14. E. Mourtaga, A. Sharieh, and M. Abdallah, "Speaker independent Quranic recognizer based on maximum likelihood linear regression," in Proceedings of world academy of science, engineering and technology, 2007, vol. 36: Citeseer, pp. 61-67. find
  15. H. Franco, L. Neumeyer, M. Ramos, and H. Bratt, "Automatic detection of phone-level mispronunciation for language learning," in Sixth European Conference on Speech Communication and Technology, 1999. find
  16. S. M. Witt, "Use of speech recognition in computer-assisted language learning," 1999. find
  17. H. Tabbal, W. El Falou, and B. Monla, "Analysis and implementation of a" Quranic" verses delimitation system in audio files using speech recognition techniques," in 2006 2nd International Conference on Information & Communication Technologies, 2006, vol. 2: IEEE, pp. 2979-2984. find
  18. R.-J. Raja-Yusof, F. Grine, N. J. Ibrahim, M. Y. I. Idris, Z. Razak, and N. N. A. Rahman, "Automated tajweed checking rules engine for Quranic learning," Multicultural Education & Technology Journal, 2013. find
  19. A. Muhammad, Z. ul Qayyum, S. Tanveer, A. Martinez-Enriquez, and A. Z. Syed, "E-hafiz: Intelligent system to help Muslims in recitation and memorization of Quran," Life Science Journal, vol. 9, no. 1, pp. 534-541, 2012. find
  20. F. A. A. R. Shafaf Ibrahim, Zaaba Ahmad, "Automatic Tajweed Rules Recognition usingk-Nearest Neighbour (k-NN)," International Journal of Recent Technology and Engineering (IJRTE), vol. Volume-8, no. Issue-2S11, 2019. find
  21. Vetterli, M., & Herley, C. (1992). Wavelets and filter banks: Theory and design. IEEE transactions on signal processing, 40(ARTICLE), 2207-2232. find
  22. B. E. Boser, I. M. Guyon, and V. N. Vapnik. A training algorithm for optimal margin classifiers. In Fish Annual Workshop on Computational Laming Theory, pages 144- 152. ACM, June 1992. find
  23. Scholkopf, B., & Smola, A. J. (2018). Learning with kernels: support vector machines, regularization, optimization, and beyond. Adaptive Computation and Machine Learning series. find
  24. Alargrami, A. M., & Eljazzar, M. M. (2020, October). Imam: Word Embedding Model for Islamic Arabic NLP. In 2020 2nd Novel Intelligent and Leading Emerging Sciences Conference (NILES) (pp. 520-524). IEEE. © 2020 By AIRCC Publishing Corporation. This article is published under the Creative Commons Attribution (CC BY) license. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/alagrami-2020-smartajweed-automatic-recognition-arabic-quranic/