---
title: "Vocal Pathologies Detection and Mispronounced Phonemes"
description: "The paper targets Arabic speech assessment. 17 references. Among them: 6."
authors: ["Naim Terbeh", "Mounir Zrigui"]
year: 2016
venue: "Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)"
source: "https://aclanthology.org/L16-1334/"
licence: "CC BY-NC 4.0 (ELRA, LREC proceedings)"
licence_url: "https://creativecommons.org/licenses/by-nc/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/terbeh-2016-vocal-pathologies-detection-mispronounced-phonemes/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/terbeh-2016-vocal-pathologies-detection-mispronounced-phonemes/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2016](https://md.research.elamur.ai/free/en/papers/2016/page/4/) / Vocal Pathologies Detection and Mispronounced Phonemes Identification: Case of Arabic Continuous Speech

# Vocal Pathologies Detection and Mispronounced Phonemes Identification: Case of Arabic Continuous Speech

**Authors:** Naim Terbeh and [Mounir Zrigui](https://md.research.elamur.ai/free/en/authors/mounir-zrigui/) · **Year:** 2016 · **Venue:** [Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)](https://md.research.elamur.ai/free/en/venues/lrec/) ([LREC 2016](https://md.research.elamur.ai/free/en/venues/lrec/2016/)) · **Publisher:** [European Language Resources Association (ELRA)](https://md.research.elamur.ai/free/en/publishers/european-language-resources-association-elra/) · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/L16-1334/) · [PDF](https://aclanthology.org/L16-1334.pdf) · [OpenAlex](https://openalex.org/W2577691310)

**Citations:** 6 citations (OpenAlex, 2026-10-14)

## Summary
The paper targets Arabic speech assessment. It decides whether a continuous utterance is healthy or pathological, and for pathological speech it finds which phonemes the speaker cannot produce and what replaces them. The intended users are native speakers with language disabilities and learners of Arabic.

Each speaker is described by a "phonetic model": a vector giving the occurrence probability of every bi-phoneme. The vector is built from forced-alignment transcriptions made with Sphinx_align and a trained Arabic acoustic model. A reference model is the average of models from several healthy sub-corpora. A new speaker's model is compared to it by the angle between the vectors. The decision threshold comes from the spread of angles among the healthy sub-corpora. The paper also says an ANN performs the final classification.

On 100 test sequences recorded with speech therapists, the overall classification rate is 97%, and 100% after combining ambiguous recordings. Phoneme identification is reported at 100%. The approach is non-invasive, unlike laryngoscopy-style diagnosis, and could support speech-correction or Arabic-learning tools.

## Key points
- **Data:** a 6-hour corpus of healthy and pathological speech, native and non-native, was used to train the acoustic model. An 11-hour healthy native corpus, split into five sub-corpora, produced the reference phonetic model.
- **Threshold:** β = Max + |Avg − δ| over the pairwise angles between the healthy sub-corpus models. A speaker whose angle θ to the reference is at most β counts as healthy.
- **Test set:** 100 sequences from 100 speakers aged 13–49: 60 pathological, 20 healthy and 20 from French non-native speakers.
- **Detection rates:** 100% for healthy speech, 95% for native pathological speech, 100% for non-native speech, and 97% overall.
- **Mispronounced phonemes:** a phoneme is flagged when bi-phonemes containing it are absent from the speaker's model but present in the reference. The count of such bi-phonemes is normalised by 57 combinations, based on 29 consonants.
  - The most frequent problem phonemes were غ, خ and ص for natives, and ظ, ض and ذ for non-natives.

## Interesting ideas
- **Self-calibrated threshold:** the healthy/abnormal cut-off is derived from how much independent clean sub-corpora differ from each other, so no hand-set value is needed.
- **Absence as evidence:** a unit that never appears where the reference expects it, aggregated over all the pairs it should occur in, pinpoints a systematic gap rather than random noise.
- **Finding substitutes through probability mass:** the probability lost by a missing unit reappears as an excess on whatever replaces it, which reveals the substitution pair.
- **Coverage depends on sample length:** short recordings may lack many combinations, which makes decisions unreliable, so the authors pool ambiguous short samples before deciding.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (learn): split a gold treebank into chunks and measure pairwise cosine angles between their deprel-bigram distributions. Batches of silver data that fall outside that natural spread can then be flagged for review.
- **I2** (tokens): list POS tags or morphological features that the tagger never emits on a domain although the reference distribution predicts them. This gives an explainable blind-spot report.
- **I3** (tree): compare relation distributions of parser output against gold to see which deprel absorbs a missing one, such as obl taking over iobj. Such pairs could seed repair rules or UD dialect-conversion checks.
- **I4** (learn): distribution-based evaluations and negative controls should enforce a minimum coverage, or pool small samples, before declaring a model or text anomalous.

*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

- [University of Sfax](https://md.research.elamur.ai/free/en/universities/university-of-sfax/) — University, Sfax, Tunisia · 4 works · 23 citations

Also on the byline: [University of Monastir](https://ror.org/00nhtcg76).

📚 Shelves (8): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 9,086 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 8,889 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 9,296 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 9,206 works · [💭 Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/) · 798 works · [💭 Experimental and Cognitive Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/experimental-and-cognitive-psychology/) · 463 works · [🩺 Medicine](https://md.research.elamur.ai/free/en/shelves/medicine/) · 80 works · [🩺 Physiology](https://md.research.elamur.ai/free/en/shelves/medicine/physiology/) · 11 works

🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 1,484 works

## Cite

Cite as: Naim Terbeh and Mounir Zrigui. 2016. Vocal Pathologies Detection and Mispronounced Phonemes Identification: Case of Arabic Continuous Speech. Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16).

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

```bibtex
@inproceedings{terbeh2016vocal,
  title = {{Vocal Pathologies Detection and Mispronounced Phonemes Identification: Case of Arabic Continuous Speech}},
  author = {Naim Terbeh and Mounir Zrigui},
  booktitle = {Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)},
  year = {2016},
  doi = {10.63317/39zc5bpnapue},
  url = {https://doi.org/10.63317/39zc5bpnapue}
}
```

- [← previous Supervised Syntax-based Alignment between English Sentences and Abstract Meaning Representation Graphs](https://md.research.elamur.ai/free/en/papers/chu-2016-supervised-syntax-based-alignment-between/)
- [next → Al Qamus al Muhit, a Medieval Arabic Lexicon in LMF](https://md.research.elamur.ai/free/en/papers/nahli-2016-al-qamus-al-muhit-medieval/)
- [↑ back to papers of 2016](https://md.research.elamur.ai/free/en/papers/2016/page/4/)

## References

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

1. Zouaghi, A., Zrigui, M., Antoniadis, G. (2008) Automatic Understanding of Spontaneous Arabic Speech-A Numerical Model. TAL, 49(1), pp. 14--166. [find](https://scholar.google.com/scholar?q=Zouaghi%2C+A%2E%2C+Zrigui%2C+M%2E%2C+Antoniadis%2C+G%2E+%282008%29+Automatic+Understanding+of+Spontaneous+Arabic+Speech%2DA+Numerical+Model%2E+TAL%2C+49%281%29%2C+pp%2E+14%2D%2D166%2E)
2. Majidnezhad, V., Kheidorov, I. (2013). An ANN-based Method for Detecting Vocal Fold Pathology. International Journal of Computer Applications, 62(7). [↗](https://doi.org/10.5120/10089-4722)
3. Majidnezhad, V., Kheidorov, I. (2012). A HMM-Based Method for Vocal Fold Pathology Diagnosis. International Journal of Computer Science Issues, 9 (6-2). [find](https://scholar.google.com/scholar?q=Majidnezhad%2C+V%2E%2C+Kheidorov%2C+I%2E+%282012%29%2E+A+HMM%2DBased+Method+for+Vocal+Fold+Pathology+Diagnosis%2E+International+Journal+of+Computer+Science+Issues%2C+9+%286%2D2%29%2E)
4. Kukharchik, P., Martynov, D., Kheidorov, I., Kotov, O. (2007). Vocal Fold Pathology Detection using Modified Wavelet-like Features and Support Vector Machines. European Signal Processing Conference. [↗](https://doi.org/10.5281/zenodo.40658)
5. Terbeh, N., Labidi, M., Zrigui M. (2013). Automatic speech correction: A step to speech recognition for people with disabilities. ICTA. Terbeh, N., Zrigui. [find](https://scholar.google.com/scholar?q=Terbeh%2C+N%2E%2C+Labidi%2C+M%2E%2C+Zrigui+M%2E+%282013%29%2E+Automatic+speech+correction%3A+A+step+to+speech+recognition+for+people+with+disabilities%2E+ICTA%2E+Terbeh%2C+N%2E%2C+Zrigui%2E)
6. M. (2014). Vers la Correction Automatique de la Parole Arabe. Citala. [find](https://scholar.google.com/scholar?q=M%2E+%282014%29%2E+Vers+la+Correction+Automatique+de+la+Parole+Arabe%2E+Citala%2E)
7. Basheer, I.A., Hajmeer, M. (2000). Artificial neural networks: fundamentals, computing, design, and application. Journal of Microbiological Methods, 43, pp. 3--31. [↗](https://doi.org/10.1016/s0167-7012(00)00201-3)
8. Lee, K.Y., Cha, Y.T., Park, J.H. (1992). SHORT-TERM LOAD FORECASTING USING ANARTIFICIAL NEURAL NETWORK. Transactions on Power Systems, 7(1), pp. 124--132. [find](https://scholar.google.com/scholar?q=Lee%2C+K%2EY%2E%2C+Cha%2C+Y%2ET%2E%2C+Park%2C+J%2EH%2E+%281992%29%2E+SHORT%2DTERM+LOAD+FORECASTING+USING+ANARTIFICIAL+NEURAL+NETWORK%2E+Transactions+on+Power+Systems%2C+7%281%29%2C+pp%2E+124%2D%2D132%2E)
9. Little, M., McSharry, P., Moroz, I., Roberts, S. (2006). Nonlinear, Biophysically-Informed Speech Pathology Detection. ICASSP. Ittichaichareon, C., Suksri, S., Yingthawornsuk. [find](https://scholar.google.com/scholar?q=Little%2C+M%2E%2C+McSharry%2C+P%2E%2C+Moroz%2C+I%2E%2C+Roberts%2C+S%2E+%282006%29%2E+Nonlinear%2C+Biophysically%2DInformed+Speech+Pathology+Detection%2E+ICASSP%2E+Ittichaichareon%2C+C%2E%2C+Suksri%2C+S%2E%2C+Yingthawornsuk%2E)
10. T. (2012). Speech Recognition using MFCC. International Conference on Computer Graphics, Simulation and Modeling. [find](https://scholar.google.com/scholar?q=T%2E+%282012%29%2E+Speech+Recognition+using+MFCC%2E+International+Conference+on+Computer+Graphics%2C+Simulation+and+Modeling%2E)
11. Archaux, C., Laanaya, H., Martin, A., Khenchaf, A. (2004). An SVM based Churn Detector in Prepaid Mobile Telephony. [find](https://scholar.google.com/scholar?q=Archaux%2C+C%2E%2C+Laanaya%2C+H%2E%2C+Martin%2C+A%2E%2C+Khenchaf%2C+A%2E+%282004%29%2E+An+SVM+based+Churn+Detector+in+Prepaid+Mobile+Telephony%2E)
12. Alonso, J.B., Leon, J.D., Alonso, I., Ferrer, M.A. (2001). Automatic Detection of Pathologies in the Voice by HOS Based Parameters. EURASIP Journal on Applied Signal Processing, 4, pp. 275-284. [↗](https://doi.org/10.1155/s1110865701000336)
13. Adnene, C., Lamia, B. (2003). Analysis of Pathological Voices by Speech Processing. Signal Processing and Its Applications, 1(1), pp. 365-367. [find](https://scholar.google.com/scholar?q=Adnene%2C+C%2E%2C+Lamia%2C+B%2E+%282003%29%2E+Analysis+of+Pathological+Voices+by+Speech+Processing%2E+Signal+Processing+and+Its+Applications%2C+1%281%29%2C+pp%2E+365%2D367%2E)
14. Jolliffe, I.T. (2008). Principal Component Analysis. Book, 2nd Edition. Bréhilin, L., Gascuel, O. Modèles de Markov caches et apprentissage de sequences. [find](https://scholar.google.com/scholar?q=Jolliffe%2C+I%2ET%2E+%282008%29%2E+Principal+Component+Analysis%2E+Book%2C+2nd+Edition%2E+Br%C3%A9hilin%2C+L%2E%2C+Gascuel%2C+O%2E+Mod%C3%A8les+de+Markov+caches+et+apprentissage+de+sequences%2E)
15. Patane, G., Russo, M. (2001). The enhanced LBG Algorithm. Neural networks, 14 (9). [↗](https://doi.org/10.1016/s0893-6080(01)00104-6)
16. Bürki, A, Gendrot, C, Gravier, G, Linarès, G, Fougeron, C. (2008). Alignement automatique et analyse phonétique: comparaison de différents systèmes pour l’analyse du schwa. TAL, 49(3), pp. 1--33. [find](https://scholar.google.com/scholar?q=B%C3%BCrki%2C+A%2C+Gendrot%2C+C%2C+Gravier%2C+G%2C+Linar%C3%A8s%2C+G%2C+Fougeron%2C+C%2E+%282008%29%2E+Alignement+automatique+et+analyse+phon%C3%A9tique%3A+comparaison+de+diff%C3%A9rents+syst%C3%A8mes+pour+l%E2%80%99analyse+du+schwa%2E+TAL%2C+49%283%29%2C+pp%2E+1%2D%2D33%2E)
17. Maraoui, M., Zrigui, M., Antoniadis G. (2012). Use of NLP Tools in CALL System for Arabic. Int. J. Comput. Proc. Oriental Lang. 24(2), pp. 153—166. [↗](https://doi.org/10.1142/s1793840612400107)

*Updated 24 October 2026*