---
title: "Voice Conversion Improves Cross-Domain — Abdullah 2025"
description: "Arabic dialect identification (ADI) systems are often used to filter dialectal speech during data collection, but they tend to fail when the audio comes…"
authors: ["Badr M. Abdullah", "Matthew Baas", "Bernd Mobius", "Dietrich Klakow"]
year: 2025
venue: "arXiv:2505.24713"
source: "https://arxiv.org/abs/2505.24713"
doi: "10.48550/arxiv.2505.24713"
arxiv: "2505.24713"
licence: "CC BY-NC-ND 4.0 (arXiv)"
licence_url: "http://creativecommons.org/licenses/by-nc-nd/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/abdullah-2025-voice-conversion-improves-cross-domain/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/abdullah-2025-voice-conversion-improves-cross-domain/)

[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/page/4/) / Voice Conversion Improves Cross-Domain Robustness for Spoken Arabic Dialect Identification

# Voice Conversion Improves Cross-Domain Robustness for Spoken Arabic Dialect Identification

**Authors:** Badr M. Abdullah, Matthew Baas, Bernd Mobius, [Dietrich Klakow](https://md.research.elamur.ai/free/en/authors/dietrich-klakow/) · **Year:** 2025 · **Venue:** arXiv:2505.24713 · **Licence:** [CC BY-NC-ND 4.0 (arXiv)](http://creativecommons.org/licenses/by-nc-nd/4.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2505.24713) · [PDF](https://arxiv.org/pdf/2505.24713) · [DOI 10.48550/arxiv.2505.24713](https://doi.org/10.48550/arxiv.2505.24713) · [arXiv 2505.24713](https://arxiv.org/abs/2505.24713) · [OpenAlex](https://openalex.org/W4415433053)

**Citations:** 1 citations (openalex-snapshot · 2026-10-14) · 6 in our library, of the 140884 papers whose bibliographies we read

## Summary
Arabic dialect identification (ADI) systems are often used to filter dialectal speech during data collection, but they tend to fail when the audio comes from a different source than the training material. The authors fine-tune MMS, a 300M-parameter wav2vec2 model, on the MGB-3 ADI-5 broadcast corpus, which covers MSA plus Gulf, Levantine, Maghrebi and Egyptian Arabic. They enlarge the training set with copies of each utterance re-voiced by voice conversion, so the dialectal content stays the same while the apparent speaker changes. To test robustness, they build MADIS-5, a manually labelled benchmark of radio, TEDx, TV drama and theatre recordings, none of which the model sees during training. With four target voices, average cross-domain accuracy rises from about 60% to almost 81%, a relative gain of roughly 34%. Conventional augmentations such as noise, pitch shift or SpecAugment fall clearly behind. The authors' analysis indicates that the gain comes partly from reducing the model's reliance on speaker identity. Both the model and the benchmark are public, which supports speech technology that covers more Arabic varieties.

## Key points
- Training data is MGB-3 ADI-5, with about 14.6k segments (~53.6 h) from Aljazeera broadcasts in 5 classes (MSA, Gulf, Levantine, Maghrebi, Egyptian). The new evaluation set, MADIS-5, has 4854 utterances (~12 h) from radio.garden stations, the SARA archive on Kaggle, the Arabic portion of TEDx, and YouTube theatre plays.
- Baselines include SVMs on phone and character n-grams, an Arabic BERT on Whisper transcripts, and fine-tuned XLSR-53, XLSR-128 and MMS. MMS on natural speech alone scores 75.94% in-domain and 60.22% on the cross-domain average.
- MMS with VC using 4 voices (5×N training samples) reaches 85.32% in-domain and 80.73% on the cross-domain average, a relative gain of +34.07%. Per domain, it scores 87.86% on radio, 86.73% on TEDx, 77.85% on dramas and 70.47% on theatre. Even a single target voice (+20.77%) beats all conventional augmentations combined (5×N, +17.49%).
- The two native-speaker annotators agreed completely on the regional dialect labels. They disagreed only on 2.3% of radio segments, and only on whether a segment was MSA or dialect.
- Releases on HuggingFace are `badrex/mms-300m-arabic-dialect-identifier` and `badrex/MADIS5-spoken-arabic-dialects`, the latter under CC BY-NC 4.0.

## Interesting ideas
- Augment with a transform that keeps the label-relevant content intact but swaps a nuisance factor (here the speaker), and train on the original plus transformed data; this attacks shortcut learning directly rather than just adding generic noise. — For Mova (learn): silver sentences could be rewritten with swapped names, numbers or spelling variants while the agreed tree stays fixed, discouraging parsers from leaning on specific lexical items.
- Compare augmentations at matched training-set sizes (2×N versus 2×N), so a gain cannot be explained simply by having more data. — For Mova (learn): when testing a new silver-data or repair-rule source, a same-size control built from naive duplication or random perturbation would act as a clean negative control.
- Judge robustness only on several genres absent from training, reported separately and averaged, instead of on a held-out split of the training domain. — For Mova (tree): parser quality could be reported on unseen genres (fables, school tests, word problems) apart from the treebank's own test split, exposing domain brittleness that LAS on in-domain data hides.
- Mapping every sample onto one fixed target voice already helps, which suggests that neutralising a nuisance variable to a constant is a cheap and useful step on its own. — For Mova (tokens): normalising surface variation such as digits, quotes, casing or entity strings into canonical placeholders before tagging could reduce spurious cues without any extra annotation.
- Annotator disagreement clustered at one known fuzzy boundary (MSA versus dialect), and the authors document it rather than forcing a crisp label. — For Mova (frames): speech-act or irony labels near a continuum could be marked as borderline in the gold data, so errors there are scored differently from clear mistakes.

*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/)

## Cite

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

```bibtex
@misc{abdullah2025voice,
  title = {{Voice Conversion Improves Cross-Domain Robustness for Spoken Arabic Dialect Identification}},
  author = {Abdullah, Badr M. and Baas, Matthew and Mobius, Bernd and Klakow, Dietrich},
  year = {2025},
  doi = {10.48550/arxiv.2505.24713},
  eprint = {2505.24713},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2505.24713}
}
```

- [← previous Visual Representations of Temporal Relations between Events and Time Expressions in News Stories](https://md.research.elamur.ai/free/en/papers/amorim-2025-visual-representations-temporal-relations-between/)
- [next → Wasm: A Pipeline for Constructing Structured Arabic Interleaved Multimodal Corpora](https://md.research.elamur.ai/free/en/papers/hennara-2025-wasm-pipeline-constructing-structured-arabic/)
- [↑ back to papers of 2025](https://md.research.elamur.ai/free/en/papers/2025/page/4/)

## References

30 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. K. Brown and S. Ogilvie, Concise encyclopedia of languages of the world. Elsevier, 2010. [find](https://scholar.google.com/scholar?q=K%2E+Brown+and+S%2E+Ogilvie%2C+Concise+encyclopedia+of+languages+of+the+world%2E+Elsevier%2C+2010%2E)
2. G. Khan, M. P. Streck, and J. C. Watson, The Semitic languages: An international handbook. Walter de Gruyter, 2011, vol. 36. [find](https://scholar.google.com/scholar?q=G%2E+Khan%2C+M%2E+P%2E+Streck%2C+and+J%2E+C%2E+Watson%2C+The+Semitic+languages%3A+An+international+handbook%2E+Walter+de+Gruyter%2C+2011%2C+vol%2E+36%2E)
3. N. Y. Habash, Introduction to Arabic natural language processing. Morgan & Claypool Publishers, 2010. [find](https://scholar.google.com/scholar?q=N%2E+Y%2E+Habash%2C+Introduction+to+Arabic+natural+language+processing%2E+Morgan+%26+Claypool+Publishers%2C+2010%2E)
4. B. Talafha, K. Kadaoui, S. M. Magdy, M. Habiboullah, C. M. Chafei, A. O. El-Shangiti, H. Zayed, M. C. Tourad, R. Alhamouri, R. Assi, A. Alraeesi, H. Mohamed, F. Alwajih, A. Mohamed, A. El Mekki, E. M. B. Nagoudi, B. D. M. Saadia, H. A. Alsayadi, W. Al-Dhabyani, S. Shatnawi, Y. Ech-chammakhy, A. Makouar, Y. Berrachedi, M. Jarrar, S. Shehata, I. Berrada, and M. Abdul-Mageed, “Casablanca: Data and models for multidialectal Arabic speech recognition,” in Proceedings of EMNLP. Association for Computational Linguistics, Nov. 2024. [find](https://scholar.google.com/scholar?q=B%2E+Talafha%2C+K%2E+Kadaoui%2C+S%2E+M%2E+Magdy%2C+M%2E+Habiboullah%2C+C%2E+M%2E+Chafei%2C+A%2E+O%2E+El%2DShangiti%2C+H%2E+Zayed%2C+M%2E+C%2E+Tourad%2C+R%2E+Alhamouri%2C+R%2E+Assi%2C+A%2E+Alraeesi%2C+H%2E+Mohamed%2C+F%2E+Alwajih%2C+A%2E+Mohamed%2C+A%2E+El+Mekki%2C+E%2E+M%2E)
5. A. Waheed, B. Talafha, P. Sullivan, A. Elmadany, and M. Abdul-Mageed, “VoxArabica: A robust dialect-aware Arabic speech recognition system,” in Proceedings of ArabicNLP 2023. Singapore (Hybrid): Association for Computational Linguistics, Dec. 2023. [find](https://scholar.google.com/scholar?q=A%2E+Waheed%2C+B%2E+Talafha%2C+P%2E+Sullivan%2C+A%2E+Elmadany%2C+and+M%2E+Abdul%2DMageed%2C+%E2%80%9CVoxArabica%3A+A+robust+dialect%2Daware+Arabic+speech+recognition+system%2C%E2%80%9D+in+Proceedings+of+ArabicNLP+2023%2E+Singapore+%28Hybrid%29%3A+Assoc)
6. P. Sullivan, A. Elmadany, and M. Abdul-Mageed, “On the robustness of arabic speech dialect identification,” in Interspeech 2023, 2023, pp. 5326–5330. [find](https://scholar.google.com/scholar?q=P%2E+Sullivan%2C+A%2E+Elmadany%2C+and+M%2E+Abdul%2DMageed%2C+%E2%80%9COn+the+robustness+of+arabic+speech+dialect+identification%2C%E2%80%9D+in+Interspeech+2023%2C+2023%2C+pp%2E+5326%E2%80%935330%2E) [our card](https://md.research.elamur.ai/free/en/papers/sullivan-2023-robustness-arabic-speech-dialect-identification/)
7. 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/)
8. S. Shon, A. Ali, and J. Glass, “Convolutional neural networks and language embeddings for end-to-end dialect recognition,” arXiv preprint arXiv:1803.04567, 2018. [↗](https://arxiv.org/abs/1803.04567)
9. A. Ali, S. Shon, Y. Samih, H. Mubarak, A. Abdelali, J. Glass, S. Renals, and K. Choukri, “The mgb-5 challenge: Recognition and dialect identification of dialectal arabic speech,” in 2019 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU). IEEE, 2019, pp. 1026–1033. [find](https://scholar.google.com/scholar?q=A%2E+Ali%2C+S%2E+Shon%2C+Y%2E+Samih%2C+H%2E+Mubarak%2C+A%2E+Abdelali%2C+J%2E+Glass%2C+S%2E+Renals%2C+and+K%2E+Choukri%2C+%E2%80%9CThe+mgb%2D5+challenge%3A+Recognition+and+dialect+identification+of+dialectal+arabic+speech%2C%E2%80%9D+in+2019+IEEE+Automati)
10. W. Lin, M. Madhavi, R. K. Das, and H. Li, “Transformer-based arabic dialect identification,” in 2020 International Conference on Asian Language Processing (IALP). IEEE, 2020, pp. 192–196. [find](https://scholar.google.com/scholar?q=W%2E+Lin%2C+M%2E+Madhavi%2C+R%2E+K%2E+Das%2C+and+H%2E+Li%2C+%E2%80%9CTransformer%2Dbased+arabic+dialect+identification%2C%E2%80%9D+in+2020+International+Conference+on+Asian+Language+Processing+%28IALP%29%2E+IEEE%2C+2020%2C+pp%2E+192%E2%80%93196%2E)
11. A. Kulkarni and H. Aldarmaki, “Yet another model for arabic dialect identification,” in Proceedings of ArabicNLP 2023, 2023, pp. 435–440. [find](https://scholar.google.com/scholar?q=A%2E+Kulkarni+and+H%2E+Aldarmaki%2C+%E2%80%9CYet+another+model+for+arabic+dialect+identification%2C%E2%80%9D+in+Proceedings+of+ArabicNLP+2023%2C+2023%2C+pp%2E+435%E2%80%93440%2E)
12. S. H. Mohammadi and A. Kain, “An overview of voice conversion systems,” Speech Communication, vol. 88, pp. 65–82, 2017. [find](https://scholar.google.com/scholar?q=S%2E+H%2E+Mohammadi+and+A%2E+Kain%2C+%E2%80%9CAn+overview+of+voice+conversion+systems%2C%E2%80%9D+Speech+Communication%2C+vol%2E+88%2C+pp%2E+65%E2%80%9382%2C+2017%2E)
13. M. Baas and H. Kamper, “Voice conversion can improve asr in very low-resource settings,” in Proc. Interspeech 2022, 2022, pp. 3513–3517. [find](https://scholar.google.com/scholar?q=M%2E+Baas+and+H%2E+Kamper%2C+%E2%80%9CVoice+conversion+can+improve+asr+in+very+low%2Dresource+settings%2C%E2%80%9D+in+Proc%2E+Interspeech+2022%2C+2022%2C+pp%2E+3513%E2%80%933517%2E)
14. E. Casanova, C. Shulby, A. Korolev, A. C. Junior, A. d. S. Soares, S. Aluı́sio, and M. A. Ponti, “Asr data augmentation in low-resource settings using cross-lingual multi-speaker tts and cross-lingual voice conversion,” in Proc. Interspeech 2023, 2023, pp. 1244–1248. [find](https://scholar.google.com/scholar?q=E%2E+Casanova%2C+C%2E+Shulby%2C+A%2E+Korolev%2C+A%2E+C%2E+Junior%2C+A%2E+d%2E+S%2E+Soares%2C+S%2E+Alu%C4%B1%CC%81sio%2C+and+M%2E+A%2E+Ponti%2C+%E2%80%9CAsr+data+augmentation+in+low%2Dresource+settings+using+cross%2Dlingual+multi%2Dspeaker+tts+and+cross%2Dlingual)
15. Y. A. Wubet and K.-Y. Lian, “Voice conversion based augmentation and a hybrid cnn-lstm model for improving speaker-independent keyword recognition on limited datasets,” IEEE Access, vol. 10, pp. 89 170–89 180, 2022. [find](https://scholar.google.com/scholar?q=Y%2E+A%2E+Wubet+and+K%2E%2DY%2E+Lian%2C+%E2%80%9CVoice+conversion+based+augmentation+and+a+hybrid+cnn%2Dlstm+model+for+improving+speaker%2Dindependent+keyword+recognition+on+limited+datasets%2C%E2%80%9D+IEEE+Access%2C+vol%2E+10%2C+pp%2E+89+17)
16. M. Baas, B. van Niekerk, and H. Kamper, “Voice conversion with just nearest neighbors,” in 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+Interspeech+2023%2C+2023%2C+pp%2E+2053%E2%80%93+2057%2E)
17. K. Qian, Y. Zhang, S. Chang, X. Yang, and M. Hasegawa-Johnson, “Autovc: Zero-shot voice style transfer with only autoencoder loss,” in International Conference on Machine Learning. PMLR, 2019, pp. 5210–5219. [find](https://scholar.google.com/scholar?q=K%2E+Qian%2C+Y%2E+Zhang%2C+S%2E+Chang%2C+X%2E+Yang%2C+and+M%2E+Hasegawa%2DJohnson%2C+%E2%80%9CAutovc%3A+Zero%2Dshot+voice+style+transfer+with+only+autoencoder+loss%2C%E2%80%9D+in+International+Conference+on+Machine+Learning%2E+PMLR%2C+2019%2C+pp%2E+521)
18. A. Polyak, Y. Adi, J. Copet, E. Kharitonov, K. Lakhotia, W.-N. Hsu, A. Mohamed, and E. Dupoux, “Speech resynthesis from discrete disentangled self-supervised representations,” in Interspeech 2021, 2021, pp. 3615–3619. [find](https://scholar.google.com/scholar?q=A%2E+Polyak%2C+Y%2E+Adi%2C+J%2E+Copet%2C+E%2E+Kharitonov%2C+K%2E+Lakhotia%2C+W%2E%2DN%2E+Hsu%2C+A%2E+Mohamed%2C+and+E%2E+Dupoux%2C+%E2%80%9CSpeech+resynthesis+from+discrete+disentangled+self%2Dsupervised+representations%2C%E2%80%9D+in+Interspeech+2021%2C+202)
19. J. Kong, J. Kim, and J. Bae, “Hifi-gan: Generative adversarial networks for efficient and high fidelity speech synthesis,” Advances in neural information processing systems, vol. 33, pp. 17 022– 17 033, 2020. [find](https://scholar.google.com/scholar?q=J%2E+Kong%2C+J%2E+Kim%2C+and+J%2E+Bae%2C+%E2%80%9CHifi%2Dgan%3A+Generative+adversarial+networks+for+efficient+and+high+fidelity+speech+synthesis%2C%E2%80%9D+Advances+in+neural+information+processing+systems%2C+vol%2E+33%2C+pp%2E+17+022%E2%80%93+17+03)
20. D. S. Park, W. Chan, Y. Zhang, C.-C. Chiu, B. Zoph, E. D. Cubuk, and Q. V. Le, “Specaugment: A simple data augmentation method for automatic speech recognition,” Interspeech 2019, p. 2613, 2019. [find](https://scholar.google.com/scholar?q=D%2E+S%2E+Park%2C+W%2E+Chan%2C+Y%2E+Zhang%2C+C%2E%2DC%2E+Chiu%2C+B%2E+Zoph%2C+E%2E+D%2E+Cubuk%2C+and+Q%2E+V%2E+Le%2C+%E2%80%9CSpecaugment%3A+A+simple+data+augmentation+method+for+automatic+speech+recognition%2C%E2%80%9D+Interspeech+2019%2C+p%2E+2613%2C+2019%2E) [our card](https://md.research.elamur.ai/free/en/papers/park-2019-specaugment-simple-data-augmentation-method/)
21. P. Foley, M. Wiesner, B. Odoom, L. P. Garcia Perera, K. Murray, and P. Koehn, “Where are you from? geolocating speech and applications to language identification,” in Proceedings of NAACL. Association for Computational Linguistics, Jun. 2024. [find](https://scholar.google.com/scholar?q=P%2E+Foley%2C+M%2E+Wiesner%2C+B%2E+Odoom%2C+L%2E+P%2E+Garcia+Perera%2C+K%2E+Murray%2C+and+P%2E+Koehn%2C+%E2%80%9CWhere+are+you+from%3F+geolocating+speech+and+applications+to+language+identification%2C%E2%80%9D+in+Proceedings+of+NAACL%2E+Association)
22. E. Salesky, M. Wiesner, J. Bremerman, R. Cattoni, M. Negri, M. Turchi, D. W. Oard, and M. Post, “Multilingual tedx corpus for speech recognition and translation,” in Proceedings of Interspeech, 2021. [find](https://scholar.google.com/scholar?q=E%2E+Salesky%2C+M%2E+Wiesner%2C+J%2E+Bremerman%2C+R%2E+Cattoni%2C+M%2E+Negri%2C+M%2E+Turchi%2C+D%2E+W%2E+Oard%2C+and+M%2E+Post%2C+%E2%80%9CMultilingual+tedx+corpus+for+speech+recognition+and+translation%2C%E2%80%9D+in+Proceedings+of+Interspeech%2C+2021%2E)
23. A. Keleg, W. Magdy, and S. Goldwater, “Estimating the level of dialectness predicts inter-annotator agreement in multi-dialect Arabic datasets,” in Proceedings of ACL, Bangkok, Thailand, Aug. 2024. [find](https://scholar.google.com/scholar?q=A%2E+Keleg%2C+W%2E+Magdy%2C+and+S%2E+Goldwater%2C+%E2%80%9CEstimating+the+level+of+dialectness+predicts+inter%2Dannotator+agreement+in+multi%2Ddialect+Arabic+datasets%2C%E2%80%9D+in+Proceedings+of+ACL%2C+Bangkok%2C+Thailand%2C+Aug%2E+2024%2E)
24. A. Keleg, S. Goldwater, and W. Magdy, “ALDi: Quantifying the Arabic level of dialectness of text,” in Proceedings of EMNLP, Singapore, Dec. 2023. [find](https://scholar.google.com/scholar?q=A%2E+Keleg%2C+S%2E+Goldwater%2C+and+W%2E+Magdy%2C+%E2%80%9CALDi%3A+Quantifying+the+Arabic+level+of+dialectness+of+text%2C%E2%80%9D+in+Proceedings+of+EMNLP%2C+Singapore%2C+Dec%2E+2023%2E)
25. M. Medvedeva, M. Kroon, and B. Plank, “When sparse traditional models outperform dense neural networks: the curious case of discriminating between similar languages,” in Proceedings of the Fourth Workshop on NLP for Similar Languages, Varieties and Dialects (VarDial), 2017, pp. 156–163. [find](https://scholar.google.com/scholar?q=M%2E+Medvedeva%2C+M%2E+Kroon%2C+and+B%2E+Plank%2C+%E2%80%9CWhen+sparse+traditional+models+outperform+dense+neural+networks%3A+the+curious+case+of+discriminating+between+similar+languages%2C%E2%80%9D+in+Proceedings+of+the+Fourth+Work)
26. G. Inoue, B. Alhafni, N. Baimukan, H. Bouamor, and N. Habash, “The interplay of variant, size, and task type in Arabic pre-trained language models,” in Proceedings of the Sixth Arabic Natural Language Processing Workshop. Kyiv, Ukraine (Online): Association for Computational Linguistics, Apr. 2021. [find](https://scholar.google.com/scholar?q=G%2E+Inoue%2C+B%2E+Alhafni%2C+N%2E+Baimukan%2C+H%2E+Bouamor%2C+and+N%2E+Habash%2C+%E2%80%9CThe+interplay+of+variant%2C+size%2C+and+task+type+in+Arabic+pre%2Dtrained+language+models%2C%E2%80%9D+in+Proceedings+of+the+Sixth+Arabic+Natural+Language) [our card](https://md.research.elamur.ai/free/en/papers/inoue-2021-interplay-variant-size-task-type/)
27. A. Babu, C. Wang, A. Tjandra, K. Lakhotia, Q. Xu, N. Goyal, K. Singh, P. von Platen, Y. Saraf, J. Pino et al., “XLSR-53: Universal cross-lingual speech representations,” in Proc. Interspeech, 2021, pp. 3429–3433. [find](https://scholar.google.com/scholar?q=A%2E+Babu%2C+C%2E+Wang%2C+A%2E+Tjandra%2C+K%2E+Lakhotia%2C+Q%2E+Xu%2C+N%2E+Goyal%2C+K%2E+Singh%2C+P%2E+von+Platen%2C+Y%2E+Saraf%2C+J%2E+Pino+et+al%2E%2C+%E2%80%9CXLSR%2D53%3A+Universal+cross%2Dlingual+speech+representations%2C%E2%80%9D+in+Proc%2E+Interspeech%2C+2021%2C+pp)
28. A. Baevski, W.-N. Hsu, Q. Xu, A. Babu, J. Gu, and A. Conneau, “XLS-R: Self-supervised cross-lingual speech representation learning at scale,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 30, pp. 2634–2650, 2022. [find](https://scholar.google.com/scholar?q=A%2E+Baevski%2C+W%2E%2DN%2E+Hsu%2C+Q%2E+Xu%2C+A%2E+Babu%2C+J%2E+Gu%2C+and+A%2E+Conneau%2C+%E2%80%9CXLS%2DR%3A+Self%2Dsupervised+cross%2Dlingual+speech+representation+learning+at+scale%2C%E2%80%9D+IEEE%2FACM+Transactions+on+Audio%2C+Speech%2C+and+Language+Proce) [our card](https://md.research.elamur.ai/free/en/papers/babu-2021-xls-r-self-supervised-cross/)
29. V. Pratap, A. Tjandra, B. Shi, P. Tomasello, A. Babu, S. Kundu, A. Elkahky, Z. Ni, A. Vyas, M. Fazel-Zarandi et al., “Scaling speech technology to 1,000+ languages,” Journal of Machine Learning Research, vol. 25, no. 97, pp. 1–52, 2024. [find](https://scholar.google.com/scholar?q=V%2E+Pratap%2C+A%2E+Tjandra%2C+B%2E+Shi%2C+P%2E+Tomasello%2C+A%2E+Babu%2C+S%2E+Kundu%2C+A%2E+Elkahky%2C+Z%2E+Ni%2C+A%2E+Vyas%2C+M%2E+Fazel%2DZarandi+et+al%2E%2C+%E2%80%9CScaling+speech+technology+to+1%2C000%2B+languages%2C%E2%80%9D+Journal+of+Machine+Learning+Resear) [our card](https://md.research.elamur.ai/free/en/papers/pratap-2023-scaling-speech-technology-1-000/)
30. D. Snyder, G. Chen, and D. Povey, “MUSAN: A Music, Speech, and Noise Corpus,” 2015, arXiv:1510.08484v1. [↗](https://arxiv.org/abs/1510.08484)