---
title: "UTD-CRSS Submission for MGB-3 Arabic Dialect Identification:"
description: "The paper presents UTD-CRSS's submission for the ADI (Arabic Dialect… 76.94% accuracy. 30 references, cited by 3 papers here."
authors: ["Ahmet E. Bulut", "Qian Zhang", "Chunlei Zhang", "Fahimeh Bahmaninezhad", "John H. L. Hansen"]
year: 2017
venue: "arXiv:1710.00113"
source: "https://arxiv.org/abs/1710.00113"
doi: "10.48550/arxiv.1710.00113"
arxiv: "1710.00113"
licence: "arXiv — non-exclusive license to distribute"
licence_url: "http://arxiv.org/licenses/nonexclusive-distrib/1.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/bulut-2017-utd-crss-submission-mgb-3/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/bulut-2017-utd-crss-submission-mgb-3/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2017](https://md.research.elamur.ai/free/en/papers/2017/page/5/) / UTD-CRSS Submission for MGB-3 Arabic Dialect Identification: Front-end and Back-end Advancements on Broadcast Speech

# UTD-CRSS Submission for MGB-3 Arabic Dialect Identification: Front-end and Back-end Advancements on Broadcast Speech

**Authors:** Ahmet E. Bulut, [Qian Zhang](https://md.research.elamur.ai/free/en/authors/qian-zhang/), Chunlei Zhang, Fahimeh Bahmaninezhad and John H. L. Hansen · **Year:** 2017 · **Venue:** [arXiv:1710.00113](https://md.research.elamur.ai/free/en/venues/ieee-automatic-speech-recognition-and-understanding-workshop-asru/) ([ASRU 2017](https://md.research.elamur.ai/free/en/venues/ieee-automatic-speech-recognition-and-understanding-workshop-asru/2017/)) · **Licence:** [arXiv — non-exclusive license to distribute](http://arxiv.org/licenses/nonexclusive-distrib/1.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/1710.00113) · [PDF](https://arxiv.org/pdf/1710.00113) · [DOI 10.48550/arxiv.1710.00113](https://doi.org/10.48550/arxiv.1710.00113) · [arXiv 1710.00113](https://arxiv.org/abs/1710.00113) · [OpenAlex](https://openalex.org/W2761207003)

**Citations:** 9 citations (OpenAlex, 2026-10-14) · [3](https://md.research.elamur.ai/free/en/papers/bulut-2017-utd-crss-submission-mgb-3/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper presents UTD-CRSS's submission for the ADI (Arabic Dialect Identification) subtask within the MGB-3 challenge. The task is to label broadcast speech from Al-Jazeera as one of five varieties: Egyptian, Gulf, Levantine, North African, or Modern Standard Arabic. The team built several individual systems and fused them.

- **Lexical branch:** an SVM over word features computed from ASR transcripts.
- **Acoustic branch:** i-Vectors derived from three front-ends (MFCCs, the organizers' bottleneck features (BNF), and unsupervised bottleneck features (UBNF)). These were scored either by a Gaussian back-end (GB) or by a semi-supervised GAN classifier that can also learn from unlabeled data.

The strongest submitted run reached 76.94% accuracy. It augmented training with a randomly selected portion of the development set. After a post-evaluation fix, accuracy rose to 79.76%, which the authors report as the best published test result for the challenge. The work shows that combining different feature types and classifiers pays off for distinguishing closely related dialects. It also shows that unlabeled audio and transcript-free bottleneck features can help when labeled data is scarce.

## Key points
- The task has five classes: EGY, GLF, LEV, NOR and MSA, drawn from Al-Jazeera TV programs. The organizers supplied audio, ASR transcripts and BNF i-Vectors.
- The final fusion has five systems:
  - unigram TF + SVM (LibShortText) on ASR text;
  - MFCC i-Vectors + GB;
  - MFCC i-Vectors + GANs;
  - BNF i-Vectors + GB;
  - UBNF i-Vectors + GANs.
- For the text systems, the authors explored 8 combinations of stop-word removal, stemming, and unigram vs. bigram. They tested BIN, TF and TF-IDF features, keeping only uni- and bigrams because data was limited.
- UBNF training needs no transcribed corpus. GMM mixture indices replace forced-aligned senone labels as the network's targets.
- Accuracy was 76.94% for the contrastive submission (dev-data augmentation) and 79.76% after the post-evaluation correction. The authors claim GANs had not been used as a DID back-end before.

## Interesting ideas
- The GAN classifier gets an extra K+1 "generated" class. Real labeled samples train the K-way supervised loss, while unlabeled real samples and generator outputs train the real-vs-fake part, so unlabeled data still shapes the class boundaries.
- Unsupervised bottleneck features use cluster indices from an unsupervised mixture model as stand-in targets for a supervised network. This yields phonetic-like representations without any transcription.
- Lexical and acoustic systems fail in different ways, and fusing them works better than any single one.
- The Gaussian back-end gives each class its own mean but shares one covariance, so scoring reduces to a transparent linear function of the input vector.
- Preprocessing choices (stop-words, stemming, n-gram order) were tested as a full factorial grid rather than picked by intuition.

## For Mova

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

- **I1** (learn): Unannotated sentences could be added to parser or role-classifier training by including an explicit "synthetic/corrupted" class, sharpening decisions without extra gold labels.
- **I2** (tokens): Cluster IDs over unlabeled English text could serve as pseudo-targets for pretraining tagger or morphology features before gold UD tags are applied.
- **I3** (learn): This supports keeping silver data from several genuinely different witness parsers, such as the perceptron parser and gparse, rather than variants of one model, so their agreement is more informative.
- **I4** (tree): A shared-covariance Gaussian scorer could cheaply and explainably detect which treebank convention or English variety an input follows before UD dialect conversion.
- **I5** (answer): Normalization options for school-test or fable QA readers could be ablated the same way, crossing every option, to show which ones actually change answer accuracy.

*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

- [The University of Texas at Dallas](https://md.research.elamur.ai/free/en/universities/the-university-of-texas-at-dallas/) — University, Richardson, United States · 61 works · 5,614 citations

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [🔊 Signal Processing](https://md.research.elamur.ai/free/en/shelves/computer-science/signal-processing/) · 452 works

🧭 Topics (2): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works · [🎙️ Speech & spoken language](https://md.research.elamur.ai/free/en/topics/speech/) · 480 works

## Cite

Cite as: Ahmet E. Bulut, Qian Zhang, Chunlei Zhang, Fahimeh Bahmaninezhad and John H. L. Hansen. 2017. UTD-CRSS Submission for MGB-3 Arabic Dialect Identification: Front-end and Back-end Advancements on Broadcast Speech. arXiv:1710.00113.

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

```bibtex
@misc{bulut2017utd,
  title = {{UTD-CRSS Submission for MGB-3 Arabic Dialect Identification: Front-end and Back-end Advancements on Broadcast Speech}},
  author = {Ahmet E. Bulut, Qian Zhang, Chunlei Zhang, Fahimeh Bahmaninezhad and John H. L. Hansen},
  year = {2017},
  doi = {10.48550/arxiv.1710.00113},
  eprint = {1710.00113},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1710.00113}
}
```

- [← previous Standard and nonstandard lexicon in aviation English: A corpus linguistic study](https://md.research.elamur.ai/free/en/papers/ferrer-2017-standard-nonstandard-lexicon-aviation-english/)
- [next → A New Error Annotation for Dyslexic texts in Arabic](https://md.research.elamur.ai/free/en/papers/alamri-2017-new-error-annotation-dyslexic-texts/)
- [↑ back to papers of 2017](https://md.research.elamur.ai/free/en/papers/2017/page/5/)

## References

30 references, 12 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Y. Lei and J. H. L. Hansen, “Dialect Classification via Text-independent Training and Testing for Arabic, Spanish, and Chinese,” IEEE Transactions on Audio, Speech, and Language Processing, vol. 19, no. 1, pp. 85–96, Jan 2011. [↗](https://doi.org/10.1109/tasl.2010.2045184)
2. V. Gupta and P. Mermelstein, “Effects of Speaker Accent on the Performance of a Speaker-independent, Isolated-word Recognizer,” The Journal of the Acoustical Society of America, vol. 71, no. 6, pp. 1581–1587, 1982. [↗](https://doi.org/10.1121/1.387812)
3. C. Huang, T. Chen, S. Li, E. Chang, and J. Zhou, “Analysis of Speaker Variability,” in EUROSPEECH 2001 - SCANDINAVIA - 7th European Conference on Speech Communication and Technology, 2001, pp. 1377–1380. [find](https://scholar.google.com/scholar?q=C%2E+Huang%2C+T%2E+Chen%2C+S%2E+Li%2C+E%2E+Chang%2C+and+J%2E+Zhou%2C+%E2%80%9CAnalysis+of+Speaker+Variability%2C%E2%80%9D+in+EUROSPEECH+2001+%2D+SCANDINAVIA+%2D+7th+European+Conference+on+Speech+Communication+and+Technology%2C+2001%2C+pp%2E+1377%E2%80%9313)
4. V. Diakoloukas, V. Digalakis, L. Neumeyer, and J. Kaja, “Development of Dialect-specific Speech Recognizers using Adaptation Methods,” in Acoustics, Speech, and Signal Processing, 1997. ICASSP-97., 1997 IEEE International Conference on. IEEE, 1997, vol. 2, pp. 1455– 1458. [find](https://scholar.google.com/scholar?q=V%2E+Diakoloukas%2C+V%2E+Digalakis%2C+L%2E+Neumeyer%2C+and+J%2E+Kaja%2C+%E2%80%9CDevelopment+of+Dialect%2Dspecific+Speech+Recognizers+using+Adaptation+Methods%2C%E2%80%9D+in+Acoustics%2C+Speech%2C+and+Signal+Processing%2C+1997%2E+ICASSP%2D97%2E%2C+19)
5. J. J. Humphries and P. C. Woodland, “The use of Accent-specific Pronunciation Dictionaries in Acoustic Model Training,” in Acoustics, Speech, and Signal Processing, 1998. Proceedings of the 1998 IEEE International Conference on. IEEE, 1998, vol. 1, pp. 317–320. [find](https://scholar.google.com/scholar?q=J%2E+J%2E+Humphries+and+P%2E+C%2E+Woodland%2C+%E2%80%9CThe+use+of+Accent%2Dspecific+Pronunciation+Dictionaries+in+Acoustic+Model+Training%2C%E2%80%9D+in+Acoustics%2C+Speech%2C+and+Signal+Processing%2C+1998%2E+Proceedings+of+the+1998+IEEE+)
6. M. Liu, B. Xu, T. Hunng, Y. Deng, and C. Li, “Mandarin Accent Adaptation Based on Context-independent/Context-dependent Pronunciation Modeling,” in Acoustics, Speech, and Signal Processing, 2000. ICASSP’00. Proceedings of the 2000 IEEE International Conference on. IEEE, 2000, vol. 2, pp. II1025– II1028. [find](https://scholar.google.com/scholar?q=M%2E+Liu%2C+B%2E+Xu%2C+T%2E+Hunng%2C+Y%2E+Deng%2C+and+C%2E+Li%2C+%E2%80%9CMandarin+Accent+Adaptation+Based+on+Context%2Dindependent%2FContext%2Ddependent+Pronunciation+Modeling%2C%E2%80%9D+in+Acoustics%2C+Speech%2C+and+Signal+Processing%2C+2000%2E+ICAS)
7. W. Ward, H. Krech, X. Yu, K. Herold, G. Figgs, A. Ikeno, D. Jurafsky, and W. Byrne, “Lexicon Adaptation for LVCSR: Speaker Idiosyncracies, Non-native Speakers, and Pronunciation Choice,” in ISCA Tutorial and Research Workshop (ITRW) on Pronunciation Modeling and Lexicon Adaptation for Spoken Language Technology, 2002. [find](https://scholar.google.com/scholar?q=W%2E+Ward%2C+H%2E+Krech%2C+X%2E+Yu%2C+K%2E+Herold%2C+G%2E+Figgs%2C+A%2E+Ikeno%2C+D%2E+Jurafsky%2C+and+W%2E+Byrne%2C+%E2%80%9CLexicon+Adaptation+for+LVCSR%3A+Speaker+Idiosyncracies%2C+Non%2Dnative+Speakers%2C+and+Pronunciation+Choice%2C%E2%80%9D+in+ISCA+Tutor)
8. H. Soltau, L. Mangu, and F. Biadsy, “From Modern Standard Arabic to Levantine ASR: Leveraging GALE for Dialects,” in Automatic Speech Recognition and Understanding (ASRU), 2017 IEEE workshop on. IEEE, 2011, pp. 266–271. [find](https://scholar.google.com/scholar?q=H%2E+Soltau%2C+L%2E+Mangu%2C+and+F%2E+Biadsy%2C+%E2%80%9CFrom+Modern+Standard+Arabic+to+Levantine+ASR%3A+Leveraging+GALE+for+Dialects%2C%E2%80%9D+in+Automatic+Speech+Recognition+and+Understanding+%28ASRU%29%2C+2017+IEEE+workshop+on%2E+IEEE%2C)
9. D. Chiang, M. T. Diab, N. Habash, O. Rambow, and S. Shareef, “[Parsing Arabic Dialects](https://md.research.elamur.ai/free/en/papers/chiang-2006-parsing-arabic-dialects/).,” in The European Chapter of the Association for Computational Linguistics (EACL), 2006.
10. A. M. Ali, N. Dehak, P. Cardinal, S. Khurana, S. H. Yella, J. R. Glass, P. Bell, and S. Renals, “[Automatic Dialect Detection in Arabic Broadcast Speech](https://md.research.elamur.ai/free/en/papers/ali-2015-automatic-dialect-detection-arabic-broadcast/),” in Interspeech 2016, 17th Annual Conference of the International Speech Communication Association, San Francisco, CA, USA, September 8-12, 2016, pp. 2934–2938.
11. O. F. Zaidan and C. Callison-Burch, “[Arabic Dialect Identification](https://md.research.elamur.ai/free/en/papers/zaidan-2014-arabic-dialect-identification/),” Comput. Linguist., vol. 40, no. 1, pp. 171–202, Mar. 2014. [↗](https://doi.org/10.1162/coli_a_00169)
12. F. Huang, “[Improved Arabic Dialect Classification with Social Media Data](https://md.research.elamur.ai/free/en/papers/huang-2015-improved-arabic-dialect-classification-social/).,” in Conference on Empirical Methods in Natural Language Processing (EMNLP), 2015, pp. 2118–2126.
13. F. Biadsy, J. Hirschberg, and N. Habash, “[Spoken Arabic Dialect Identification Using Phonotactic Modeling](https://md.research.elamur.ai/free/en/papers/biadsy-2009-spoken-arabic-dialect-identification-phonotactic/),” in Proceedings of the EACL 2009 Workshop on Computational Approaches to Semitic Languages, Stroudsburg, PA, USA, 2009, Semitic ’09, pp. 53–61, Association for Computational Linguistics.
14. J. H. L. Hansen and G. Liu, “Unsupervised Accent Classification for Deep Data Fusion of Accent and Language Information,” Speech Communication, vol. 78, pp. 19– 33, 2016. [↗](https://doi.org/10.1016/j.specom.2015.12.004)
15. H. Bořil, A. Sangwan, and J. H. L. Hansen, “Arabic dialect identification - ’is the secret in the silence?’ and other observations,” in Proc. Interspeech, 2012. [find](https://scholar.google.com/scholar?q=H%2E+Bor%CC%8Cil%2C+A%2E+Sangwan%2C+and+J%2E+H%2E+L%2E+Hansen%2C+%E2%80%9CArabic+dialect+identification+%2D+%E2%80%99is+the+secret+in+the+silence%3F%E2%80%99+and+other+observations%2C%E2%80%9D+in+Proc%2E+Interspeech%2C+2012%2E)
16. M. Zampieri, S. Malmasi, N. Ljubešić, P. Nakov, A. Ali, J. Tiedemann, Y. Scherrer, and N. Aepli, “[Findings of the VarDial Evaluation Campaign 2017](https://md.research.elamur.ai/free/en/papers/zampieri-2017-findings-vardial-evaluation-campaign/),” in In Proceedings of the VarDial Workshop, 2017.
17. Y. Belinkov and J. Glass, “[A Character-level Convolutional Neural Network for Distinguishing Similar Languages and Dialects](https://md.research.elamur.ai/free/en/papers/belinkov-2016-character-level-convolutional-neural-network/),” In Proceedings of the VarDial Workshop, 2016. [↗](https://doi.org/10.48550/arxiv.1609.07568)
18. C. Guggilla, “Discrimination between Similar Languages, Varieties and Dialects using CNN-and LSTM-based Deep Neural Networks,” In Proceedings of the VarDial Workshop, 2016. — [Discrimination between Similar Languages, Varieties and Dialects using CNN- and LSTM-based Deep Neural Networks](https://md.research.elamur.ai/free/en/papers/guggilla-2016-discrimination-between-similar-languages-varieties/)
19. S. Malmasi, M. Zampieri, N. Ljubešic, P. Nakov, A. Ali, and J. Tiedemann, “[Discriminating between Similar Languages and Arabic Dialect Identification: A Report on the Third DSL Shared Task](https://md.research.elamur.ai/free/en/papers/malmasi-2016-discriminating-between-similar-languages/),” In Proceedings of the VarDial Workshop, 2016.
20. H. Yu, C. Ho, Y. Juan, and C. Lin, “LibShortText: A Library for Short-text Classification and Analysis,” Rapport interne, Department of Computer Science, National Taiwan University. Software available at [http://www.csie](http://www.csie). ntu. edu. tw/˜ cjlin/libshorttext, 2013. [↗](http://www.csie)
21. G. Liu, T. Hasen, H. Boril, and J. H. L. Hansen, “An Investigation on Back-end for Speaker Recognition in Multi-session Enrollment,” in Acoustics, Speech, and Signal Processing, 2013. Proceedings of the 2013 IEEE International Conference on, Vancouver, Canada, 2013. [find](https://scholar.google.com/scholar?q=G%2E+Liu%2C+T%2E+Hasen%2C+H%2E+Boril%2C+and+J%2E+H%2E+L%2E+Hansen%2C+%E2%80%9CAn+Investigation+on+Back%2Dend+for+Speaker+Recognition+in+Multi%2Dsession+Enrollment%2C%E2%80%9D+in+Acoustics%2C+Speech%2C+and+Signal+Processing%2C+2013%2E+Proceedings+of+t)
22. T. Salimans, I. Goodfellow, W. Zaremba, V. Cheung, A. Radford, and X. Chen, “[Improved Techniques for Training GANs](https://md.research.elamur.ai/free/en/papers/salimans-2016-improved-techniques-training-gans/),” in Advances in Neural Information Processing Systems, 2016, pp. 2234–2242. [↗](https://doi.org/10.48550/arxiv.1606.03498)
23. Q. Zhang and J. H. L. Hansen, “Dialect Recognition Based on Unsupervised Bottleneck Features,” in Interspeech 2017, 18th Annual Conference of the International Speech Communication Association, Stockholm, Sweden, Aug. 20-24, 2017. [find](https://scholar.google.com/scholar?q=Q%2E+Zhang+and+J%2E+H%2E+L%2E+Hansen%2C+%E2%80%9CDialect+Recognition+Based+on+Unsupervised+Bottleneck+Features%2C%E2%80%9D+in+Interspeech+2017%2C+18th+Annual+Conference+of+the+International+Speech+Communication+Association%2C+Stockh)
24. A. Ali, S. Vogel, and S. Renals, “[Speech Recognition Challenge in the Wild: Arabic MGB-3](https://md.research.elamur.ai/free/en/papers/ali-2017-speech-recognition-challenge-wild-arabic/),” in Automatic Speech Recognition and Understanding (ASRU), 2017 IEEE workshop on. IEEE, 2017.
25. S. Greenberg, “On the Origins of Speech Intelligibility in the Real World,” in Robust Speech Recognition for Unknown Communication Channels, 1997. [find](https://scholar.google.com/scholar?q=S%2E+Greenberg%2C+%E2%80%9COn+the+Origins+of+Speech+Intelligibility+in+the+Real+World%2C%E2%80%9D+in+Robust+Speech+Recognition+for+Unknown+Communication+Channels%2C+1997%2E)
26. N. Dehak, P. J. Kenny, R. Dehak, P. Dumouchel, and P. Ouellet, “Front-end Factor Analysis for Speaker Verification,” IEEE Transactions on Audio, Speech, and Language Processing, vol. 19, no. 4, pp. 788–798, 2011. [↗](https://doi.org/10.1109/tasl.2010.2064307)
27. I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio, “Generative Adversarial Nets,” in Advances in Neural Information Processing Systems, 2014, pp. 2672–2680. [find](https://scholar.google.com/scholar?q=I%2E+Goodfellow%2C+J%2E+Pouget%2DAbadie%2C+M%2E+Mirza%2C+B%2E+Xu%2C+D%2E+Warde%2DFarley%2C+S%2E+Ozair%2C+A%2E+Courville%2C+and+Y%2E+Bengio%2C+%E2%80%9CGenerative+Adversarial+Nets%2C%E2%80%9D+in+Advances+in+Neural+Information+Processing+Systems%2C+2014%2C+pp%2E)
28. C. Zhang, Q. Zhang, and J. H. L. Hansen, “Semi-supervised Learning with Generative Adversarial Networks for Arabic Dialect Identification,” Submitted to IEEE Signal Processing Letters, 2017. [find](https://scholar.google.com/scholar?q=C%2E+Zhang%2C+Q%2E+Zhang%2C+and+J%2E+H%2E+L%2E+Hansen%2C+%E2%80%9CSemi%2Dsupervised+Learning+with+Generative+Adversarial+Networks+for+Arabic+Dialect+Identification%2C%E2%80%9D+Submitted+to+IEEE+Signal+Processing+Letters%2C+2017%2E)
29. 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](https://md.research.elamur.ai/free/en/papers/ali-2016-mgb-2-challenge-arabic-multi/),” in Spoken Language Technology Workshop (SLT), 2016 IEEE. IEEE, 2016, pp. 279–284.
30. N. Brümmer, “FoCal Multi-class: Toolkit for Evaluation, Fusion and Calibration of Multi-class Recognition Scores (Tutorial and User Manual),” Software available at [http://sites.google.com/site/nikobrummer/focalmulticlass](http://sites.google.com/site/nikobrummer/focalmulticlass), 2007. [↗](http://sites.google.com/site/nikobrummer/focalmulticlass)

*Updated 1 November 2026*