Elamur Research⌕
☰
home / papers / 2021 / QASR: QCRI Aljazeera Speech Resource A Large Scale Annotated Arabic Speech Corpus

QASR: QCRI Aljazeera Speech Resource A Large Scale Annotated Arabic Speech Corpus

Paper · 2021 · Hamdy Mubarak, Amir Hussein, Shammur Absar Chowdhury et al.

Summary

The authors tackle the lack of a large Arabic resource that connects speech with NLP-level annotation and covers both Modern Standard Arabic and dialects. They built QASR from about 11 years (2004–2015) of the Aljazeera archive. It contains roughly 2,000 hours of 16kHz audio from over 4,000 episodes. The original human transcripts had no timing and were not verbatim, so the authors ran an ASR system over each episode. They then aligned its output to the transcript with a relaxed local alignment and cut the result into short, linguistically sensible segments that keep punctuation. Speaker names were cleaned and normalized across the corpus, and gender and country labels were added where possible. They also released a 130M-word text collection for language modelling. Finally, they enriched the MGB-2 testset with speaker, dialect and named-entity layers. End-to-end ASR trained on QASR reaches word error rates competitive with MGB-2. The paper also provides first baselines for Arabic punctuation restoration and for NER on speech transcripts. The resource matters because it lets researchers study spoken Arabic jointly through acoustics and text.

Key points

  • QASR has about 2,041 hours, 1.6M segments and 14.3M words. Segments average about 4 seconds and 9 words. The corpus has 27,977 speaker turns, corresponding to 11,092 unique speakers.
  • Speaker gender is given for 2,000 speakers, covering about 82% of segments. Gender and country labels are only assigned to Arabic speakers with at least 20 segments, because foreign speakers are dubbed.
  • Alignment uses the Smith–Waterman algorithm, modified so that two words match approximately if their Levenshtein distance is at most half the transcript word's length. Each segment carries AWD, GWER and WMER quality scores.
  • The MGB-2 testset (about 70K words) was manually annotated for PER, LOC, ORG and OTH entities. Farasa NER drops from 84.3 F1 on ANERcorp to 69.8 F1 on these transcripts.
  • Text-based and acoustic dialect ID agree on the MSA/DA decision 67.5% of the time. Most errors come from confusing neighbouring countries, such as Syria vs Lebanon or Tunisia vs Algeria.

Interesting ideas

  • I1. Imperfect, non-verbatim transcripts are matched to machine output with local alignment that tolerates near-miss words, so noisy pairs still align reliably without requiring exact identity.
  • I2. Every released segment comes with its own match-error and duration scores, so users can pick their own quality threshold instead of receiving one fixed filtered set.
  • I3. Speaker names were normalized across the corpus and flagged as either a specific unique person or a generic role like an unnamed protester, which separates identity linking from descriptive labels.
  • I4. The same decision was made independently from two modalities (text and acoustics), and their agreement rate was reported as a diagnostic of how hard the labelling task is.
  • I5. An off-the-shelf tool was tested on clean news and on a shifted domain (speech transcripts) with the same label scheme, exposing a sizable gap that a single benchmark would hide.

For Mova

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

  • I1 (back): The text → graph → text compressor could align regenerated sentences to the source with a similar fuzzy local alignment, scoring which spans survived the round trip.
  • I2 (learn): Silver parses could ship with per-sentence witness-agreement scores, letting training runs choose strict or lenient subsets and test how noise level affects the parser.
  • I3 (world): Entities in a text's world could carry a similar specific-vs-generic flag, so coreference does not merge "a farmer" in one tale with a named farmer elsewhere.
  • I4 (learn): Measuring agreement between independent analysers, such as the perceptron parser vs gparse, gives a cheap difficulty signal and flags sentences needing an LLM teacher.
  • I5 (answer): Readers could be evaluated on paired in-domain and shifted sets, for example school texts vs fables, to show where frame and world building degrade.

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

🎓Hamad bin Khalifa UniversityDoha · Qatar71 works · 12,942 citations

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 9,086 worksArtificial Intelligence · 8,889 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 1,484 worksSpeech & spoken language · 233 works

Cite

Cite as Hamdy Mubarak, Amir Hussein, Shammur Absar Chowdhury and Ahmed Ali. 2021. QASR: QCRI Aljazeera Speech Resource A Large Scale Annotated Arabic Speech Corpus. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2021.acl-long.177

@inproceedings{mubarak2021qasr,
  title = {{QASR: QCRI Aljazeera Speech Resource A Large Scale Annotated Arabic Speech Corpus}},
  author = {Hamdy Mubarak, Amir Hussein, Shammur Absar Chowdhury and Ahmed Ali},
  booktitle = {Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)},
  year = {2021},
  doi = {10.18653/v1/2021.acl-long.177},
  url = {https://doi.org/10.18653/v1/2021.acl-long.177}
}

References

52 references, 15 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ahmed Abdelali, Kareem Darwish, Nadir Durrani, and Hamdy Mubarak. 2016. Farasa: A fast and furious segmenter for Arabic. In Proceedings of the 2016 conference of the North American chapter of the association for computational linguistics: Demonstrations, pages 11–16.
  2. Ahmed Abdelali, Hamdy Mubarak, Younes Samih, Sabit Hassan, and Kareem Darwish. 2020. Arabic dialect identification in the wild. arXiv preprint arXiv:2005.06557. ↗
  3. Muhammad Abdul-Mageed, Chiyu Zhang, Houda Bouamor, and Nizar Habash. 2020. Nadi 2020: The first nuanced Arabic dialect identification shared task. arXiv preprint arXiv:2010.11334. ↗
  4. Tanvirul Alam, Akib Khan, and Firoj Alam. 2020. Punctuation restoration using transformer models for resource-rich and-poor languages. In Proceedings of the Sixth Workshop on Noisy User-generated Text (W-NUT 2020), pages 132–142. find
  5. Ahmed Ali, Peter Bell, James Glass, Yacine Messaoui, Hamdy Mubarak, Steve Renals, and Yifan Zhang. 2016. The MGB-2 challenge: Arabic multi-dialect broadcast media recognition. In 2016 IEEE Spoken Language Technology Workshop (SLT), pages 279– 284. IEEE.
  6. Ahmed Ali, Shammur Chowdhury, Mohamed Afify, Wassim El-Hajj, Hazem Hajj, Mourad Abbas, Amir Hussein, Nada Ghneim, Mohammad Abushariah, and Assal Alqudah. 2021. Connecting Arabs: bridging the gap in dialectal speech recognition. Communications of the ACM, 64(4):124–129. ↗
  7. Ahmed Ali, Suwon Shon, Younes Samih, Hamdy Mubarak, Ahmed Abdelali, James Glass, Steve Renals, and Khalid Choukri. 2019. The MGB-5 challenge: Recognition and dialect identification of dialectal Arabic speech. In IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), pages 1026–1033. IEEE. find
  8. Ahmed Ali, Stephan Vogel, and Steve Renals. 2017. Speech recognition challenge in the wild: Arabic MGB-3. In IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), pages 316– 322. IEEE.
  9. Wissam Antoun, Fady Baly, and Hazem Hajj. 2020. Arabert: Transformer-based model for Arabic language understanding. In LREC 2020 Workshop Language Resources and Evaluation Conference 11–16 May 2020, page 9. ↗
  10. Rosana Ardila, Megan Branson, Kelly Davis, Michael Henretty, Michael Kohler, Josh Meyer, Reuben Morais, Lindsay Saunders, Francis M Tyers, and Gregor Weber. 2019. Common voice: A massively-multilingual speech corpus. arXiv preprint arXiv:1912.06670. ↗
  11. Peter Bell, Mark JF Gales, Thomas Hain, Jonathan Kilgour, Pierre Lanchantin, Xunying Liu, Andrew Mc-Parland, Steve Renals, Oscar Saz, Mirjam Wester, et al. 2015. The MGB challenge: Evaluating multi-genre broadcast media recognition. In 2015 IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU), pages 687–693. IEEE. find
  12. Yassine Benajiba and Paolo Rosso. 2008. Arabic named entity recognition using conditional random fields. In Proc. of Workshop on HLT & NLP within the Arabic World, LREC, volume 8, pages 143–153. Citeseer. find
  13. Norbert Braunschweiler, Mark JF Gales, and Sabine Buchholz. 2010. Lightly supervised recognition for automatic alignment of large coherent speech recordings. In Eleventh Annual Conference of the International Speech Communication Association. find
  14. Chung-Cheng Chiu, Tara N Sainath, Yonghui Wu, Rohit Prabhavalkar, Patrick Nguyen, Zhifeng Chen, Anjuli Kannan, Ron J Weiss, Kanishka Rao, Ekaterina Gonina, et al. 2018. State-of-the-art speech recognition with sequence-to-sequence models. In 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 4774–4778. IEEE. find
  15. Shammur A Chowdhury, Ahmed Ali, Suwon Shon, and James Glass. 2020a. What does an end-to-end dialect identification model learn about non-dialectal information? Proc. Interspeech 2020, pages 462– 466. find
  16. Shammur A Chowdhury, Younes Samih, Mohamed Eldesouki, and Ahmed Ali. 2020b. Effects of dialectal code-switching on speech modules: A study using Egyptian Arabic broadcast speech. Proc. Interspeech. find
  17. Shammur Absar Chowdhury, Amir Hussein, Ahmed Abdelali, and Ahmed Ali. 2021. Towards one model to rule all: Multilingual strategy for dialectal code-switching Arabic Asr. arXiv:2105.14779. ↗
  18. Shammur Absar Chowdhury, Evgeny Stepanov, Morena Danieli, and Giuseppe Riccardi. 2017. Functions of silences towards information flow in spoken conversation. In Proceedings of the Workshop on Speech-Centric Natural Language Processing, pages 1–9. find
  19. Shammur Absar Chowdhury, Evgeny A Stepanov, Morena Danieli, and Giuseppe Riccardi. 2019. Automatic classification of speech overlaps: Feature representation and algorithms. Computer Speech & Language, 55:145–167. ↗
  20. Christopher Cieri, David Miller, and Kevin Walker. 2004. The fisher corpus: a resource for the next generations of speech-to-text. In LREC, volume 4, pages 69–71.
  21. Kareem Darwish. 2013. Named entity recognition using cross-lingual resources: Arabic as an example. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1558–1567. find
  22. Chenpeng Du, Hao Li, Yizhou Lu, Lan Wang, and Yanmin Qian. 2021. Data augmentation for end-to-end code-switching speech recognition. In 2021 IEEE Spoken Language Technology Workshop (SLT), pages 194–200. IEEE. find
  23. Pilar Oplustil Gallegos, Jennifer Williams, Joanna Rownicka, and Simon King. 2020. An unsupervised method to select a speaker subset from large multi-speaker speech synthesis datasets. Proc. Interspeech 2020, pages 1758–1762. find
  24. Björn Gambäck and Amitava Das. 2016. Comparing the Level of Code-Switching in Corpora. In LREC. find
  25. Pegah Ghahremani, Bagher BabaAli, Daniel Povey, Korbinian Riedhammer, Jan Trmal, and Sanjeev Khudanpur. 2014. A pitch extraction algorithm tuned for automatic speech recognition. In IEEE international conference on acoustics, speech and signal processing (ICASSP), pages 2494–2498. IEEE. find
  26. Sabit Hassan, Hamdy Mubarak, Ahmed Abdelali, and Kareem Darwish. 2021. Asad: Arabic social media analytics and understanding. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations, pages 113–118. find
  27. Amir Hussein, Shinji Watanabe, and Ahmed Ali. 2021. Arabic speech recognition by end-to-end, modular systems and human. arXiv:2101.08454. arXiv preprint Sameer Khurana and Ahmed Ali. QCRI advanced transcription system (QATS) for the Arabic multi-dialect broadcast media recognition: MGB-2 challenge. Training, 1200(2214):370K. ↗
  28. Tom Ko, Vijayaditya Peddinti, Daniel Povey, and Sanjeev Khudanpur. 2015. Audio augmentation for speech recognition. In Sixteenth Annual Conference of the International Speech Communication Association. find
  29. Xinxing Li and Edward Lin. 2020. A 43 language multilingual punctuation prediction neural network model. Proc. Interspeech 2020, pages 1067–1071. find
  30. Hamdy Mubarak, Ahmed Abdelali, Hassan Sajjad, Younes Samih, and Kareem Darwish. 2019. Highly effective Arabic diacritization using sequence to sequence modeling. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 2390–2395. find
  31. Hamdy Mubarak and Kareem Darwish. 2014. Automatic correction of Arabic text: A cascaded approach. In Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP), pages 132–136. find
  32. Hamdy Mubarak, Kareem Darwish, and Ahmed Abdelali. 2015. Qcri@ qalb-2015 shared task: Correction of Arabic text for native and non-native speakers’ errors. In Proceedings of the Second Workshop on Arabic Natural Language Processing, pages 150–154. find
  33. Arsha Nagrani, Joon Son Chung, and Andrew Zisserman. 2017. Voxceleb: a large-scale speaker identification dataset. arXiv preprint arXiv:1706.08612. ↗
  34. Joseph Olive, Caitlin Christianson, and John McCary. 2011. Handbook of natural language processing and machine translation: DARPA global autonomous language exploitation. Springer Science & Business Media. ↗
  35. David S Pallett. 2003. A look at nist’s benchmark asr tests: past, present, and future. In 2003 IEEE Workshop on Automatic Speech Recognition and Understanding (IEEE Cat. No. 03EX721), pages 483–488. IEEE. find
  36. Vassil Panayotov, Guoguo Chen, Daniel Povey, and Sanjeev Khudanpur. 2015. Librispeech: an asr corpus based on public domain audio books. In 2015 IEEE international conference on acoustics, speech and signal processing (ICASSP), pages 5206–5210. IEEE. find
  37. Daniel S Park, William Chan, Yu Zhang, Chung-Cheng Chiu, Barret Zoph, Ekin D Cubuk, and Quoc V Le. 2019. Specaugment: A simple data augmentation method for automatic speech recognition. Proc. Interspeech, pages 2613–2617.
  38. Arfath Pasha, Mohamed Al-Badrashiny, Mona T Diab, Ahmed El Kholy, Ramy Eskander, Nizar Habash, Manoj Pooleery, Owen Rambow, and Ryan Roth. 2014. Madamira: A fast, comprehensive tool for morphological analysis and disambiguation of Arabic. In LREC, volume 14, pages 1094–1101. Citeseer.
  39. Sheena Christabel Pravin and M Palanivelan. 2021. A hybrid deep ensemble for speech disfluency classification. Circuits, Systems, and Signal Processing, pages 1–28. ↗
  40. Younes Samih, Mohammed Attia, Mohamed Eldesouki, Ahmed Abdelali, Hamdy Mubarak, Laura Kallmeyer, and Kareem Darwish. 2017. A neural architecture for dialectal Arabic segmentation. In Proceedings of the Third Arabic Natural Language Processing Workshop, pages 46–54.
  41. George Saon, Gakuto Kurata, Tom Sercu, Kartik Audhkhasi, Samuel Thomas, Dimitrios Dimitriadis, Xiaodong Cui, Bhuvana Ramabhadran, Michael Picheny, Lynn-Li Lim, et al. 2017. English conversational telephone speech recognition by humans and machines. arXiv preprint arXiv:1703.02136. ↗
  42. Shinji Watanabe, Takaaki Hori, Shigeki Karita, Tomoki Hayashi, Jiro Nishitoba, Yuya Unno, Nelson-Enrique Yalta Soplin, Jahn Heymann, Matthew Wiesner, Nanxin Chen, et al. 2018. Espnet: End-to-end speech processing toolkit. Proc. Interspeech, pages 2207–2211. find
  43. Jane Wottawa, Marie Tahon, Apolline Marin, and Nicolas Audibert. 2020. Towards interactive annotation for hesitation in conversational speech. In LREC 2020. find
  44. Wayne Xiong, Jasha Droppo, Xuedong Huang, Frank Seide, Mike Seltzer, Andreas Stolcke, Dong Yu, and Geoffrey Zweig. 2016. Achieving human parity in conversational speech recognition. arXiv preprint arXiv:1610.05256. ↗
  45. Suwon Shon, Ahmed Ali, and James Glass. 2017. MIT-QCRI Arabic dialect identification system for the 2017 multi-genre broadcast challenge. In 2017 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), pages 374–380. IEEE. find
  46. Suwon Shon, Ahmed Ali, Younes Samih, Hamdy Mubarak, and James Glass. 2020. Adi17: A fine-grained Arabic dialect identification dataset. In ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 8244–8248. IEEE. find
  47. Suwon Shon, Najim Dehak, Douglas Reynolds, and James Glass. 2019. Mce 2018: The 1st multi-target speaker detection and identification challenge evaluation. arXiv preprint arXiv:1904.04240. ↗
  48. Suwon Shon, Hao Tang, and James Glass. 2018. Frame-level speaker embeddings for text-independent speaker recognition and analysis of end-to-end model. In 2018 IEEE Spoken Language Technology Workshop (SLT), pages 1007–1013. IEEE. find
  49. Temple F Smith, Michael S Waterman, et al. 1981. Identification of common molecular subsequences. Journal of molecular biology, 147(1):195–197. ↗
  50. Anshuman Tripathi, Han Lu, and Hasim Sak. 2020. End-to-end multi-talker overlapping speech recognition. In ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 6129–6133. IEEE. find
  51. Jörgen Valk and Tanel Alumäe. 2020. Voxlingua107: a dataset for spoken language recognition. arXiv preprint arXiv:2011.12998. ↗
  52. Changhan Wang, Morgane Rivière, Ann Lee, Anne Wu, Chaitanya Talnikar, Daniel Haziza, Mary Williamson, Juan Pino, and Emmanuel Dupoux. 2021. Voxpopuli: A large-scale multilingual speech corpus for representation learning, semi-supervised learning and interpretation. arXiv preprint arXiv:2101.00390. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mubarak-2021-qasr-qcri-aljazeera-speech-resource/