---
title: "Saudi ASWAT — paper by Alharbi et al. (2026)"
description: "The authors present Saudi ASWAT (Arabic Speech With Annotated Transcripts), described as the first nationwide corpus of… 23 references."
authors: ["Abdullah I. Alharbi", "Afrah A. Altamimi", "Muneera Alhoshan", "Amal Almazrua", "Halah Munif Alharbi", "Bayan M. Almuqhim", "Hawra Aljasim", "Abdulrahman Alosaimy", "Yahya A. Asiri", "Abdullah Alfaifi"]
year: 2026
venue: "Proceedings of the Fifteenth Language Resources and Evaluation Conference"
source: "https://aclanthology.org/2026.lrec-1.124/"
doi: "10.63317/5dmcn5u57hg6"
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/alharbi-2026-saudi-aswat-large-scale-corpus/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/alharbi-2026-saudi-aswat-large-scale-corpus/)

[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/9/) / Saudi ASWAT: A Large-Scale Corpus of Spontaneous Saudi Arabic Speech

# Saudi ASWAT: A Large-Scale Corpus of Spontaneous Saudi Arabic Speech

**Authors:** [Abdullah I. Alharbi](https://md.research.elamur.ai/free/en/authors/abdullah-i-alharbi/), [Afrah A. Altamimi](https://md.research.elamur.ai/free/en/authors/afrah-altamimi/), [Muneera Alhoshan](https://md.research.elamur.ai/free/en/authors/muneera-alhoshan/), [Amal Almazrua](https://md.research.elamur.ai/free/en/authors/amal-almazrua/), [Halah Munif Alharbi](https://md.research.elamur.ai/free/en/authors/halah-munif-alharbi/), [Bayan M. Almuqhim](https://md.research.elamur.ai/free/en/authors/bayan-m-almuqhim/), [Hawra Aljasim](https://md.research.elamur.ai/free/en/authors/hawra-aljasim/), [Abdulrahman Alosaimy](https://md.research.elamur.ai/free/en/authors/abdulrahman-alosaimy/), [Yahya A. Asiri](https://md.research.elamur.ai/free/en/authors/yahya-a-asiri/) and Abdullah Alfaifi · **Year:** 2026 · **Venue:** [Proceedings of the Fifteenth Language Resources and Evaluation Conference](https://md.research.elamur.ai/free/en/venues/lrec/) ([LREC 2026](https://md.research.elamur.ai/free/en/venues/lrec/2026/)) · **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/2026.lrec-1.124/) · [PDF](https://aclanthology.org/2026.lrec-1.124.pdf) · [DOI 10.63317/5dmcn5u57hg6](https://doi.org/10.63317/5dmcn5u57hg6) · [OpenAlex](https://openalex.org/W7160068445)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
Most Arabic speech resources contain Modern Standard Arabic or broadcast material. Dialectal, unscripted everyday talk is therefore poorly represented, which hurts ASR, diarization and sociolinguistic work. The authors present Saudi ASWAT (Arabic Speech With Annotated Transcripts), described as the first nationwide corpus of spontaneous Saudi speech. It targets 2,500 hours of natural conversation from five regional varieties (Najdi, Eastern, Hijazi, Northern, Southern), which together contain roughly fifty-five local varieties.

Native fieldworkers collected the speech in real settings such as homes, schools, workplaces and public places, using participants' own devices. Speakers were balanced by gender and across age bands. The recordings are transcribed under a standardized, dialect-aware scheme and linked to speaker-level and clip-level metadata. A TF–IDF analysis of local varieties brings out regionally distinctive vocabulary.

The resource matters because earlier Saudi corpora are either scraped monologues without transcripts or TV-derived speech shaped by production. Saudi ASWAT offers authentic, acoustically varied, demographically stratified data. It also proposes an ethically governed protocol for building such corpora.

## Key points
- **Scale and collection:** 2,500 hours were gathered over 10 months by 55 trained field recorders. Each recorder was responsible for about 40–50 hours.
- **Coverage:** Five regional zones with about 55 local varieties. Speakers fall into five age bands, from children (5–10) to older adults (40–60), with gender balance. All participants or their guardians gave written consent.
- **Elicitation:**
  - Fieldworkers held short conversations about everyday interests, speaking the local dialect.
  - They nudged speakers to keep going when they paused.
  - With children especially, they used wordless cartoon stories.
  - Prompts drew on local culture to encourage local vocabulary.
- **Comparison with prior resources:** The earlier ASWAT dataset (Alkanhal et al., 2023) has 732 hours of mostly monologue audio from YouTube and SoundCloud, without transcripts, and only 27% of it is Saudi dialect. SADA (Alharbi et al., 2024) has 668 hours from 57 TV shows across 11 genres. Older dialect telephone corpora are typically 20–60 hours at 8 kHz.
- **Access and future work:** A sample is available to qualified researchers on request, and an online platform (falak.ksaa.gov.sa) offers search and visualization tools. Future work includes staged quality review and filling underrepresented areas such as Hail and Al-Bahah.

## Interesting ideas
- Wordless picture stories make many speakers describe the same fixed content in their own words, so variation in expression can be compared against a constant meaning.
- Per-region TF–IDF profiling is a cheap, transparent way to surface vocabulary that departs from the standard lexicon.
- The authors argue that polished broadcast data breeds models that fail on messy real input, so they deliberately chose uncontrolled devices and settings.
- Each item carries structured metadata at two levels (speaker and clip), and sampling is stratified by demographic cells.
- Collectors were native speakers who addressed participants in the local dialect, so the elicitor's own register would not push speakers toward the formal standard.

## For Mova

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

- **I1** (back): Render one world graph of a scene into many surface texts, then check whether the compressor maps all of them back to the same graph.
- **I2** (tokens): Run TF–IDF per source or genre (fables, school texts, forums) to list distinctive words that the lexicon or lemmatizer is missing, and review them before they cause tagging errors.
- **I3** (learn): Keep a held-out set of informal, unedited English (disfluent, chatty, typo-ridden) as a stress test, so silver-data training is not tuned only to clean prose.
- **I4** (learn): Tag every training and evaluation sentence with source, register and genre metadata so parser and frame errors can be sliced by subgroup instead of hidden in one averaged score.
- **I5** (learn): When an LLM teacher generates practice text or questions, match the target register (for example, tale-style for fables) so the training data is not shifted toward a formal default.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 9,085 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 8,888 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

🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 1,485 works

## Cite

Cite as: Abdullah I. Alharbi, Afrah A. Altamimi, Muneera Alhoshan, Amal Almazrua, Halah Munif Alharbi, Bayan M. Almuqhim, Hawra Aljasim, Abdulrahman Alosaimy, Yahya A. Asiri and Abdullah Alfaifi. 2026. Saudi ASWAT: A Large-Scale Corpus of Spontaneous Saudi Arabic Speech. Proceedings of the Fifteenth Language Resources and Evaluation Conference.

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

```bibtex
@inproceedings{alharbi2026saudi,
  title = {{Saudi ASWAT: A Large-Scale Corpus of Spontaneous Saudi Arabic Speech}},
  author = {Abdullah I. Alharbi, Afrah A. Altamimi, Muneera Alhoshan, Amal Almazrua, Halah Munif Alharbi, Bayan M. Almuqhim, Hawra Aljasim, Abdulrahman Alosaimy, Yahya A. Asiri and Abdullah Alfaifi},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/5dmcn5u57hg6},
  url = {https://doi.org/10.63317/5dmcn5u57hg6}
}
```

- [← previous Santham: A Curated Sanskrit–Tamil Dataset with Anvaya and Segmentation for Building and Evaluating Machine Translation](https://md.research.elamur.ai/free/en/papers/s-2026-santham-curated-sanskrit-tamil-dataset/)
- [next → SAVI: Web-based Multilayered Semantic Annotation Validation Interface](https://md.research.elamur.ai/free/en/papers/tatavolu-2026-savi-web-based-multilayered-semantic/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/9/)

## References

23 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Sadeen Alharbi, Areeb Alowisheq, Zoltán Tüske, Kareem Darwish, Abdullah Alrajeh, et al. 2024. Sada: Saudi audio dataset for arabic. [https://www.kaggle.com/datasets/sdaiancai/sada2022](https://www.kaggle.com/datasets/sdaiancai/sada2022). [↗](https://www.kaggle.com/datasets/sdaiancai/sada2022)
2. Lamya Alkanhal, Abeer Alessa, Elaf Almahmoud, and Rana Alaqil. 2023. Aswat: Arabic audio dataset for automatic speech recognition using speech-representation learning. In Proceedings of the First Arabic Natural Language Processing Conference, pages 120–127. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Lamya+Alkanhal%2C+Abeer+Alessa%2C+Elaf+Almahmoud%2C+and+Rana+Alaqil%2E+2023%2E+Aswat%3A+Arabic+audio+dataset+for+automatic+speech+recognition+using+speech%2Drepresentation+learning%2E+In+Proceedings+of+the+First+Arab)
3. Alexei Baevski, Yuhao Zhou, Abdelrahman Mohamed, and Michael Auli. 2020. [wav2vec 2.0: A framework for self-supervised learning of speech representations](https://md.research.elamur.ai/free/en/papers/baevski-2020-wav2vec-2-0-framework-self/). arXiv preprint arXiv:2006.11477. [↗](https://arxiv.org/abs/2006.11477)
4. Jeong-Uk Bang, Seung Yun, Seunghi Kim, Mu-Yeol Choi, Min-Kyu Lee, Yeojeong Kim, Dong-Hyun Kim, Jun Park, Youngjik Lee, and Sang-Hun Kim. 2020. Ksponspeech: Korean spontaneous speech corpus for automatic speech recognition. Applied Sciences, 10:6936. [↗](https://doi.org/10.3390/app10196936)
5. He Bu, Jun Du, Xu Na, Bengu Wu, and Hui Zheng. 2017. Aishell-1: An open-source mandarin speech corpus. [https://openslr.org/33](https://openslr.org/33). [↗](https://openslr.org/33)
6. Alexandra Canavan, David Graff, and George Zipperlen. 1997. Callhome egyptian arabic speech. [https://catalog.ldc.upenn.edu/LDC97S45](https://catalog.ldc.upenn.edu/LDC97S45). LDC97S45. [↗](https://catalog.ldc.upenn.edu/LDC97S45)
7. Christopher Cieri, David Miller, and Kevin Walker. 2004. Fisher english training speech part 1. [https://catalog.ldc.upenn.edu/LDC2004S13](https://catalog.ldc.upenn.edu/LDC2004S13). LDC2004S13. [↗](https://catalog.ldc.upenn.edu/LDC2004S13)
8. Charles A. Ferguson. 1959. Diglossia. WORD, 15(2):325–340. [find](https://scholar.google.com/scholar?q=Charles+A%2E+Ferguson%2E+1959%2E+Diglossia%2E+WORD%2C+15%282%29%3A325%E2%80%93340%2E)
9. S. Furui, Masanobu Nakamura, Tomohisa Ichiba, and K. Iwano. 2005. Analysis and recognition of spontaneous speech using corpus of spontaneous japanese. Speech Commun., 47:208– 219. [↗](https://doi.org/10.1016/j.specom.2005.02.010)
10. John Godfrey, Edward Holliman, and Jane McDaniel. 1997. Switchboard-1 release 2. [https://catalog.ldc.upenn.edu/LDC97S62](https://catalog.ldc.upenn.edu/LDC97S62). LDC97S62. [↗](https://catalog.ldc.upenn.edu/LDC97S62)
11. Nizar Habash, Owen Rambow, and George Kiraz. 2012. [Conventional orthography for dialectal arabic](https://md.research.elamur.ai/free/en/papers/habash-2012-conventional-orthography-dialectal-arabic/). In Proceedings of the Language Resources and Evaluation Conference (LREC), pages 1251– 1258.
12. Clive Holes. 2006. The arabic dialects of arabia. Proceedings of the Seminar for Arabian Studies, 36:25–34. [find](https://scholar.google.com/scholar?q=Clive+Holes%2E+2006%2E+The+arabic+dialects+of+arabia%2E+Proceedings+of+the+Seminar+for+Arabian+Studies%2C+36%3A25%E2%80%9334%2E)
13. LDC. 2006a. Gulf arabic conversational telephone speech. [https://catalog.ldc.upenn.edu/LDC2006S48](https://catalog.ldc.upenn.edu/LDC2006S48). LDC2006S48. [↗](https://catalog.ldc.upenn.edu/LDC2006S48)
14. LDC. 2006b. Iraqi arabic conversational telephone speech. [https://catalog.ldc.upenn.edu/LDC2006S29](https://catalog.ldc.upenn.edu/LDC2006S29). LDC2006S29. [↗](https://catalog.ldc.upenn.edu/LDC2006S29)
15. Rodrigo Lima, Sidney E. Leal, Arnaldo Candido Junior, and Sandra M. Aluísio. 2025. A large dataset of spontaneous speech with the accent spoken in são paulo for automatic speech recognition evaluation. In Intelligent Systems: 34th Brazilian Conference, BRACIS 2024, Belém Do Pará, Brazil, November 17–21, 2024, Proceedings, Part I, page 33–47, Berlin, Heidelberg. Springer-Verlag. [find](https://scholar.google.com/scholar?q=Rodrigo+Lima%2C+Sidney+E%2E+Leal%2C+Arnaldo+Candido+Junior%2C+and+Sandra+M%2E+Alu%C3%ADsio%2E+2025%2E+A+large+dataset+of+spontaneous+speech+with+the+accent+spoken+in+s%C3%A3o+paulo+for+automatic+speech+recognition+evaluation)
16. Deyi Liu and Pascale Fung. 2005. Hkust mandarin telephone speech. [https://catalog.ldc.upenn.edu/LDC2005S15](https://catalog.ldc.upenn.edu/LDC2005S15). LDC2005S15. [↗](https://catalog.ldc.upenn.edu/LDC2005S15)
17. Kikuo Maekawa. 2003. Corpus of spontaneous japanese (csj). [https://clrd.ninjal.ac.jp/csj/en/](https://clrd.ninjal.ac.jp/csj/en/). [↗](https://clrd.ninjal.ac.jp/csj/en/)
18. John Makhoul and Bassam Zawaydeh. 2005. Bbn/aub darpa babylon levantine arabic speech and transcripts. [https://catalog.ldc.upenn.edu/LDC2005S08](https://catalog.ldc.upenn.edu/LDC2005S08). LDC2005S08. [↗](https://catalog.ldc.upenn.edu/LDC2005S08)
19. Mark A. Pitt, Laura Dilley, Keith Johnson, Scott Kiesling, William Raymond, Elizabeth Hume, and Eric Fosler-Lussier. 2002. The buckeye corpus of conversational speech. [https://buckeyecorpus.osu.edu/](https://buckeyecorpus.osu.edu/). [↗](https://buckeyecorpus.osu.edu/)
20. Matt Post and John Garofolo. 2010. Fisher spanish training speech part 1. [https://catalog.ldc.upenn.edu/LDC2010S01](https://catalog.ldc.upenn.edu/LDC2010S01). LDC2010S01. [↗](https://catalog.ldc.upenn.edu/LDC2010S01)
21. Thomas Prochazka. 1988. Saudi Arabian Dialects, 1 edition. Routledge. [find](https://scholar.google.com/scholar?q=Thomas+Prochazka%2E+1988%2E+Saudi+Arabian+Dialects%2C+1+edition%2E+Routledge%2E)
22. Alec Radford, Jong Wook Kim, Christopher Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. 2023. [Robust speech recognition via large-scale weak supervision](https://md.research.elamur.ai/free/en/papers/radford-2022-robust-speech-recognition-large-scale/). arXiv preprint arXiv:2306.16217. [↗](https://arxiv.org/abs/2306.16217)
23. Wolfgang Wahlster. 2000. Verbmobil speech databases. [https://www.phonetik.uni-muenchen.de/Bas/BasVM.html](https://www.phonetik.uni-muenchen.de/Bas/BasVM.html). 10. Language Resource References [↗](https://www.phonetik.uni-muenchen.de/Bas/BasVM.html)

*Updated 24 October 2026*