---
title: "ArVoice: A Multi-Speaker Dataset for Arabic — Toyin 2025"
description: "Arabic speech synthesis lacks clean, multi-speaker training data with full diacritics."
authors: ["Hawau Olamide Toyin", "Rufael Marew", "Humaid Alblooshi", "Samar M. Magdy", "Hanan Aldarmaki"]
year: 2025
venue: "arXiv:2505.20506"
source: "https://arxiv.org/abs/2505.20506"
doi: "10.48550/arxiv.2505.20506"
arxiv: "2505.20506"
licence: "CC BY 4.0 (arXiv)"
licence_url: "http://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/toyin-2025-arvoice-multi-speaker-dataset-arabic/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/toyin-2025-arvoice-multi-speaker-dataset-arabic/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2025](https://md.research.elamur.ai/free/en/papers/2025/) / ArVoice: A Multi-Speaker Dataset for Arabic Speech Synthesis

# ArVoice: A Multi-Speaker Dataset for Arabic Speech Synthesis

**Authors:** [Hawau Olamide Toyin](https://md.research.elamur.ai/free/en/authors/hawau-olamide-toyin/), Rufael Marew, Humaid Alblooshi, [Samar M. Magdy](https://md.research.elamur.ai/free/en/authors/samar-m-magdy/), [Hanan Aldarmaki](https://md.research.elamur.ai/free/en/authors/hanan-aldarmaki/) · **Year:** 2025 · **Venue:** arXiv:2505.20506 · **Licence:** [CC BY 4.0 (arXiv)](http://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2505.20506) · [PDF](https://arxiv.org/pdf/2505.20506) · [DOI 10.48550/arxiv.2505.20506](https://doi.org/10.48550/arxiv.2505.20506) · [arXiv 2505.20506](https://arxiv.org/abs/2505.20506) · [OpenAlex](https://openalex.org/W4415433204)

**Citations:** 2 citations (openalex-snapshot · 2026-10-14) · [4](#cited-by) in our library, of the 140884 papers whose bibliographies we read

## Summary
Arabic speech synthesis lacks clean, multi-speaker training data with full diacritics. Without diacritics, the vowels are unspecified, so a TTS system has to guess how each word is pronounced. Existing Arabic speech collections were mostly built for ASR from noisy broadcast or web audio. The two open TTS corpora, ASC and ClArTTS, each have only one male speaker, and ClArTTS is Classical Arabic rather than MSA.

ArVoice fills this gap with four parts:
- new studio recordings by four voice talents reading diacritized MSA news text from the Al Jazeera portion of Tashkeela,
- two further talents reading undiacritized Khaleej Corpus text,
- a cleaned-up subset of ASC whose phonetic-style spellings were restored to standard orthography,
- parallel synthetic speech from Google commercial voices.

The full corpus totals 83.52 hours across 11 voices, of which about 10 hours are human speech from 7 speakers. The authors train open-source TTS and voice conversion baselines on it. The corpus matters because it gives MSA a multi-speaker, diacritized benchmark that also suits voice conversion, speech-based diacritization and deepfake detection.

## Key points
- Composition: 83.52 h in total, 11 voices, about 10 h of human speech from 7 speakers (roughly 9 h train, 0.86 h test). The speakers come from Egypt, Jordan, Morocco, Palestine and Syria.
- Part 1 has about 1.8K diacritized pseudo-sentences plus 200 that contain digits. Long sentences (over 50 characters) were split at commas, keeping at least 20 characters per piece. Each of the 2 male and 2 female talents read different text.
- Part 3 re-spells ASC into standard orthography, covering nunation written as a letter and the missing Alef in the article. GPT-4 made a first pass and humans corrected it. Duplicate nonsense sentences were dropped, leaving about 0.9K sentences.
- Zero-shot ASR checks used ArTST, Nvidia CTC-large and Whisper-large. WER fell from about 45–46% on the original ASC transcripts to 5.1–7.5% on the corrected ones.
- Licensing: the ASC part and the synthetic part are released under CC BY 4.0 at huggingface.co/datasets/MBZUAI/ArVoice. The professional recordings require a signed Data Usage Agreement.

## Interesting ideas
- Several independent off-the-shelf ASR models served as external judges of a transcript cleanup, and the large WER drop shown by all of them counted as evidence that the corrections were real. — For Mova (learn): a normalization or conversion change could be accepted only when several independent witness parsers agree that downstream quality improved, rather than trusting one metric.
- Nonstandard spellings were first repaired automatically by an LLM with a fixed prompt, and then every item was checked by hand. This makes a cheap proposal step followed by a trusted verification step. — For Mova (tokens): an LLM teacher tier could propose lemma or spelling normalizations for odd tokens, with only verified fixes promoted into the morphology tables.
- Transcripts were edited to match what speakers actually said, keeping repetitions and omissions. The authors found that a number like 2020 can be voiced in a different form than a rule-based verbalizer predicts. — For Mova (back): number-to-words generation should keep several valid verbalizations (e.g., "twenty twenty" vs "two thousand twenty") rather than one canonical output, and comparisons should accept any of them.
- The synthetic, parallel portion is explicitly labelled as having unguaranteed quality and is recommended only for augmentation and voice conversion. Data provenance thus limits how each subset may be used. — For Mova (learn): silver and LLM-generated data could carry a usage tag (training augmentation only, never evaluation) so that weaker sources cannot leak into test sets.

*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](https://md.research.elamur.ai/free/en/shelves/linguistics/)

## Cited by

Cited by

4 papers of our library cite it (of the 140884 whose bibliographies we read); 1 of them have a page here, 3 not yet.

- [More Data, Fewer Diacritics: Scaling Arabic TTS](https://md.research.elamur.ai/free/en/papers/musleh-2026-more-data-fewer-diacritics-scaling/)

## Cite

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

```bibtex
@misc{toyin2025arvoice,
  title = {{ArVoice: A Multi-Speaker Dataset for Arabic Speech Synthesis}},
  author = {Toyin, Hawau Olamide and Marew, Rufael and Alblooshi, Humaid and Magdy, Samar M. and Aldarmaki, Hanan},
  year = {2025},
  doi = {10.48550/arxiv.2505.20506},
  eprint = {2505.20506},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2505.20506}
}
```

- [← previous ArFake: A Robust Framework for Multi-Dialect Arabic Speech Spoofing Detection Benchmark](https://md.research.elamur.ai/free/en/papers/elsetohy-2025-arfake-robust-framework-multi-dialect/)
- [next → ArzEn-MultiGenre: An aligned parallel dataset of Egyptian Arabic song lyrics, novels, and subtitles, with English translations](https://md.research.elamur.ai/free/en/papers/al-2025-sabbagh-arzen-multigenre-aligned-parallel-dataset/)
- [↑ back to papers of 2025](https://md.research.elamur.ai/free/en/papers/2025/)

## References

21 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. A. Ali, S. Vogel, and S. Renals, “Speech recognition challenge in the wild: Arabic mgb-3,” in 2017 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU). IEEE, 2017, pp. 316–322. [find](https://scholar.google.com/scholar?q=A%2E+Ali%2C+S%2E+Vogel%2C+and+S%2E+Renals%2C+%E2%80%9CSpeech+recognition+challenge+in+the+wild%3A+Arabic+mgb%2D3%2C%E2%80%9D+in+2017+IEEE+Automatic+Speech+Recognition+and+Understanding+Workshop+%28ASRU%29%2E+IEEE%2C+2017%2C+pp%2E+316%E2%80%93322%2E) [our card](https://md.research.elamur.ai/free/en/papers/ali-2017-speech-recognition-challenge-wild-arabic/)
2. H. Mubarak, A. Hussein, S. A. Chowdhury, and A. Ali, “QASR: QCRI aljazeera speech resource a large scale annotated Arabic speech corpus,” in 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), C. Zong, F. Xia, W. Li, and R. Navigli, Eds. Online: Association for Computational Linguistics, Aug. 2021, pp. 2274–2285. [Online]. Available: https://aclanthology.org/2021.acl-long.177/ [↗](https://aclanthology.org/2021.acl-long.177/)
3. M. Al-Fetyani, M. Al-Barham, G. Abandah, A. Alsharkawi, and M. Dawas, “Masc: Massive arabic speech corpus,” in 2022 IEEE Spoken Language Technology Workshop (SLT). IEEE, 2023, pp. 1006–1013. [find](https://scholar.google.com/scholar?q=M%2E+Al%2DFetyani%2C+M%2E+Al%2DBarham%2C+G%2E+Abandah%2C+A%2E+Alsharkawi%2C+and+M%2E+Dawas%2C+%E2%80%9CMasc%3A+Massive+arabic+speech+corpus%2C%E2%80%9D+in+2022+IEEE+Spoken+Language+Technology+Workshop+%28SLT%29%2E+IEEE%2C+2023%2C+pp%2E+1006%E2%80%931013%2E)
4. H. Aldarmaki and A. Ghannam, “Diacritic recognition performance in arabic asr,” in Interspeech 2023, 2023, pp. 361–365. [find](https://scholar.google.com/scholar?q=H%2E+Aldarmaki+and+A%2E+Ghannam%2C+%E2%80%9CDiacritic+recognition+performance+in+arabic+asr%2C%E2%80%9D+in+Interspeech+2023%2C+2023%2C+pp%2E+361%E2%80%93365%2E) [our card](https://md.research.elamur.ai/free/en/papers/aldarmaki-2023-diacritic-recognition-performance-arabic-asr/)
5. S. Shatnawi, S. Alqahtani, and H. Aldarmaki, “Automatic restoration of diacritics for speech data sets,” in Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), 2024, pp. 4166–4176. [find](https://scholar.google.com/scholar?q=S%2E+Shatnawi%2C+S%2E+Alqahtani%2C+and+H%2E+Aldarmaki%2C+%E2%80%9CAutomatic+restoration+of+diacritics+for+speech+data+sets%2C%E2%80%9D+in+Proceedings+of+the+2024+Conference+of+the+North+American+Chapter+of+the+Association+for+Comp)
6. A. Kulkarni, A. Kulkarni, S. A. M. Shatnawi, and H. Aldarmaki, “Clartts: An open-source classical arabic text-to-speech corpus,” in Interspeech 2023, 2023, pp. 5511–5515. [find](https://scholar.google.com/scholar?q=A%2E+Kulkarni%2C+A%2E+Kulkarni%2C+S%2E+A%2E+M%2E+Shatnawi%2C+and+H%2E+Aldarmaki%2C+%E2%80%9CClartts%3A+An+open%2Dsource+classical+arabic+text%2Dto%2Dspeech+corpus%2C%E2%80%9D+in+Interspeech+2023%2C+2023%2C+pp%2E+5511%E2%80%935515%2E) [our card](https://md.research.elamur.ai/free/en/papers/kulkarni-2023-clartts-open-source-classical-arabic/)
7. A. W. Black, “Cmu wilderness multilingual speech dataset,” in ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2019, pp. 5971– 5975. [find](https://scholar.google.com/scholar?q=A%2E+W%2E+Black%2C+%E2%80%9CCmu+wilderness+multilingual+speech+dataset%2C%E2%80%9D+in+ICASSP+2019%2D2019+IEEE+International+Conference+on+Acoustics%2C+Speech+and+Signal+Processing+%28ICASSP%29%2E+IEEE%2C+2019%2C+pp%2E+5971%E2%80%93+5975%2E)
8. I. Hamed, N. T. Vu, and S. Abdennadher, “ArzEn: A speech corpus for code-switched Egyptian Arabic-English,” in Proceedings of the Twelfth Language Resources and Evaluation Conference, N. Calzolari, F. Béchet, P. Blache, K. Choukri, C. Cieri, T. Declerck, S. Goggi, H. Isahara, B. Maegaard, J. Mariani, H. Mazo, A. Moreno, J. Odijk, and S. Piperidis, Eds. Marseille, France: European Language Resources Association, May 2020, pp. 4237–4246. [Online]. Available: https://aclanthology.org/2020.lrec-1.523 [↗](https://aclanthology.org/2020.lrec-1.523/)
9. Y. Lin, X. Qin, G. Zhao, M. Cheng, N. Jiang, H. Wu, and M. Li, “Voxblink: A large scale speaker verification dataset on camera,” ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 10 271–10 275, 2023. [Online]. Available: [https://api.semanticscholar.org/CorpusID:260927071](https://api.semanticscholar.org/CorpusID:260927071) [↗](https://api.semanticscholar.org/CorpusID:260927071)
10. N. Halabi, “Modern standard arabic phonetics for speech synthesis,” Ph.D. dissertation, University of Southampton, July 2016. [Online]. Available: [https://eprints.soton.ac.uk/409695/](https://eprints.soton.ac.uk/409695/) [↗](https://eprints.soton.ac.uk/409695/)
11. T. Zerrouki and A. Balla, “Tashkeela: Novel corpus of arabic vocalized texts, data for auto-diacritization systems,” Data in Brief, vol. 11, 02 2017. [find](https://scholar.google.com/scholar?q=T%2E+Zerrouki+and+A%2E+Balla%2C+%E2%80%9CTashkeela%3A+Novel+corpus+of+arabic+vocalized+texts%2C+data+for+auto%2Ddiacritization+systems%2C%E2%80%9D+Data+in+Brief%2C+vol%2E+11%2C+02+2017%2E)
12. M. Abbas and K. Smaı̈li, “Comparison of topic identification methods for arabic language,” in International Conference on Recent Advances in Natural Language Processing-RANLP 2005, no. 14-17, 2005. [find](https://scholar.google.com/scholar?q=M%2E+Abbas+and+K%2E+Sma%C4%B1%CC%88li%2C+%E2%80%9CComparison+of+topic+identification+methods+for+arabic+language%2C%E2%80%9D+in+International+Conference+on+Recent+Advances+in+Natural+Language+Processing%2DRANLP+2005%2C+no%2E+14%2D17%2C+2005%2E)
13. H. O. Toyin, H. Li, and H. Aldarmaki, “STTATTS: Unified speech-to-text and text-to-speech model,” in Findings of the Association for Computational Linguistics: EMNLP 2024, Y. Al-Onaizan, M. Bansal, and Y.-N. Chen, Eds. Miami, Florida, USA: Association for Computational Linguistics, Nov. 2024, pp. 6853–6863. [Online]. Available: https://aclanthology.org/2024. findings-emnlp.401/ [↗](https://aclanthology.org/2024)
14. A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak supervision,” in International Conference on Machine Learning, 2022. [Online]. Available: [https://api.semanticscholar.org/CorpusID:252923993](https://api.semanticscholar.org/CorpusID:252923993) [↗](https://api.semanticscholar.org/CorpusID:252923993) [our card](https://md.research.elamur.ai/free/en/papers/radford-2022-robust-speech-recognition-large-scale/)
15. H. O. Toyin, A. Djanibekov, A. Kulkarni, and H. Aldarmaki, “ArTST: Arabic text and speech transformer,” in Proceedings of ArabicNLP 2023. Singapore (Hybrid): Association for Computational Linguistics, Dec. 2023, pp. 41–51. [Online]. Available: https://aclanthology.org/2023.arabicnlp-1.5/ [↗](https://aclanthology.org/2023.arabicnlp-1.5/)
16. D. Rekesh, N. R. Koluguri, S. Kriman, S. Majumdar, V. Noroozi, H. Huang, O. Hrinchuk, K. Puvvada, A. Kumar, J. Balam et al., “Fast conformer with linearly scalable attention for efficient speech recognition,” in 2023 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU). IEEE, 2023, pp. 1–8. [find](https://scholar.google.com/scholar?q=D%2E+Rekesh%2C+N%2E+R%2E+Koluguri%2C+S%2E+Kriman%2C+S%2E+Majumdar%2C+V%2E+Noroozi%2C+H%2E+Huang%2C+O%2E+Hrinchuk%2C+K%2E+Puvvada%2C+A%2E+Kumar%2C+J%2E+Balam+et+al%2E%2C+%E2%80%9CFast+conformer+with+linearly+scalable+attention+for+efficient+speech+recog)
17. J. Kim, J. Kong, and J. Son, “Conditional variational autoencoder with adversarial learning for end-to-end text-to-speech,” in International Conference on Machine Learning. PMLR, 2021, pp. 5530–5540. [find](https://scholar.google.com/scholar?q=J%2E+Kim%2C+J%2E+Kong%2C+and+J%2E+Son%2C+%E2%80%9CConditional+variational+autoencoder+with+adversarial+learning+for+end%2Dto%2Dend+text%2Dto%2Dspeech%2C%E2%80%9D+in+International+Conference+on+Machine+Learning%2E+PMLR%2C+2021%2C+pp%2E+5530%E2%80%935540%2E)
18. S. Liao, Y. Wang, T. Li, Y. Cheng, R. Zhang, R. Zhou, and Y. Xing, “Fish-Speech: Leveraging large language models for advanced multilingual Text-to-Speech synthesis,” arXiv (Cornell University), 11 2024. [Online]. Available: http://arxiv.org/abs/2411.01156 [↗](https://arxiv.org/abs/2411.01156)
19. W.-C. Huang, K. Kobayashi, and T. Toda, “Aas-vc: On the generalization ability of automatic alignment search based non-autoregressive sequence-to-sequence voice conversion,” 2023. [Online]. Available: https://arxiv.org/abs/2309.07598 [↗](https://arxiv.org/abs/2309.07598)
20. R. Yamamoto, E. Song, and J.-M. Kim, “Parallel wavegan: A fast waveform generation model based on generative adversarial networks with multi-resolution spectrogram,” in ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2020, pp. 6199–6203. [find](https://scholar.google.com/scholar?q=R%2E+Yamamoto%2C+E%2E+Song%2C+and+J%2E%2DM%2E+Kim%2C+%E2%80%9CParallel+wavegan%3A+A+fast+waveform+generation+model+based+on+generative+adversarial+networks+with+multi%2Dresolution+spectrogram%2C%E2%80%9D+in+ICASSP+2020%2D2020+IEEE+Internati)
21. M. Baas, B. van Niekerk, and H. Kamper, “Voice conversion with just nearest neighbors,” in Proc. Interspeech 2023, 2023, pp. 2053–2057. [find](https://scholar.google.com/scholar?q=M%2E+Baas%2C+B%2E+van+Niekerk%2C+and+H%2E+Kamper%2C+%E2%80%9CVoice+conversion+with+just+nearest+neighbors%2C%E2%80%9D+in+Proc%2E+Interspeech+2023%2C+2023%2C+pp%2E+2053%E2%80%932057%2E)