Elamur Research⌕
☰
home / papers / 2026 / TaLK-Corpus: A Regionally Diverse Evaluation Set for Sri Lankan Tamil Speech

TaLK-Corpus: A Regionally Diverse Evaluation Set for Sri Lankan Tamil Speech

Paper · 2026 · Adsajan Thillainathan, Nishanthini Kanthakumar, Nivethiga Rasan et al.

Summary

Sri Lankan Tamil has no benchmark for evaluating speech recognition. Existing Tamil ASR datasets mostly capture Indian Tamil, and the main Sri Lankan resource, EmoTa, targets emotion recognition. The authors present TaLK-Corpus, a small evaluation set organised by geography. It has one native speaker from each of 22 Sri Lankan districts, recorded speaking spontaneously on everyday topics. The audio was converted to 16 kHz mono WAV and split into utterances with Silero VAD. Linguists then transcribed it manually in Tamil script and added broad IPA. Each utterance comes with speaker metadata, and files are anonymised through MD5-based naming. Two multilingual models, Whisper Large-V3 and MMS, were tested zero-shot. Both scored poorly overall, with WER above 0.8, and worse than published results on Indian Tamil. Accuracy also varied widely from district to district, and common errors came from English code-switching. The resource lets researchers compare ASR systems district by district and with dialect in mind, giving a starting point for more inclusive Sri Lankan Tamil speech technology.

Key points

  • Corpus contents: 1 h 33 min of speech in 1,214 utterances from 22 speakers (12 female, 10 male), covering 22 of the country's 25 districts.
  • Annotation: Tamil-script transcripts plus IPA, generated with ThamizhiIPA-Trans and checked by a linguist for Jaffna, and written by hand for the other districts. A single CSV file holds every metadata field, recording each speaker's gender, age range, occupation, religion and district, along with the topic and duration.
  • Zero-shot WER/CER:
    • Whisper v3 reached WER 0.807 and CER 0.425.
    • facebook/mms-1b-all reached WER 0.845 and CER 0.342.
    • Published IndicVoices Tamil results are 0.784 for Whisper and 0.754 for MMS.
  • Variation by district: Whisper's WER ranges from 0.672 to 0.903. Hatton, Moneragala and Badulla score better for both models, possibly because speech there is closer to Indian Tamil.
  • Availability and limitations: released under CC BY-NC 4.0 at github.com/LTG-UoJ/TaLK-Corpus-Public. With only one speaker per district, dialect effects cannot be separated from individual speaker traits. Inconsistent spacing in the orthography also inflates WER.

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

🎓University of JaffnaJaffna · Sri Lanka7 works · 70 citations

 Fields, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksPsychology · 1,552 worksExperimental and Cognitive Psychology · 870 worksSignal Processing · 452 works

🧭 Topics 1Research subjects and languages the work is about.Speech & spoken language · 480 works

Cite

Cite as Adsajan Thillainathan, Nishanthini Kanthakumar, Nivethiga Rasan and Kengatharaiyer Sarveswaran. 2026. TaLK-Corpus: A Regionally Diverse Evaluation Set for Sri Lankan Tamil Speech. Proceedings of Speech Language Models in Low-Resource Settings: Performance, Evaluation, and Bias Analysis (SPEAKABLE) @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/5edtmzph3wu4

@inproceedings{thillainathan2026talk,
  title = {{TaLK-Corpus: A Regionally Diverse Evaluation Set for Sri Lankan Tamil Speech}},
  author = {Adsajan Thillainathan, Nishanthini Kanthakumar, Nivethiga Rasan and Kengatharaiyer Sarveswaran},
  booktitle = {Proceedings of Speech Language Models in Low-Resource Settings: Performance, Evaluation, and Bias Analysis (SPEAKABLE) @ LREC 2026},
  year = {2026},
  doi = {10.63317/5edtmzph3wu4},
  url = {https://doi.org/10.63317/5edtmzph3wu4}
}

References

21 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Harsh Ahlawat, Naveen Aggarwal, and Deepti Gupta. 2025. Automatic Speech Recognition: A survey of deep learning techniques and approaches. International Journal of Cognitive Computing in Engineering, 6:201–237. find
  2. B. Bharathi, Bharathi Raja Chakravarthi, et al. 2022. Findings of the Shared Task on Speech Recognition for Vulnerable Individuals in Tamil. https://sites.google.com/univ.jfn.ac. lk/talkcorpus/home In Proceedings of the Second Workshop on Language Technology for Equality, Diversity and Inclusion. ↗
  3. B. Bharathi, Bharathi Raja Chakravarthi, et al. 2024. Overview of the Third Shared Task on Speech Recognition for Vulnerable Individuals in Tamil. In Proceedings of the Fourth Workshop on Language Technology for Equality, Diversity, Inclusion. find
  4. Akriti Dhasmana, Aarohi Srivastava, and David Chiang. 2026. Dialect Matters: Cross-Lingual ASR Transfer for Low-Resource Indic Language Varieties. arXiv preprint arXiv:2601.04373. find
  5. P. Jain and A. Bhowmick. 2025. Comparative performance analysis of end-to-end ASR models on Indo-Aryan and Dravidian languages within India’s linguistic landscape. Journal of Audio, Speech, and Music Processing, 10(2025). find
  6. Tahir Javed, Janki Nawale, Eldho George, Sakshi Joshi, Kaushal Bhogale, Deovrat Mehendale, Ishvinder Sethi, Aparna Ananthanarayanan, Hafsah Faquih, Pratiti Palit, Sneha Ravishankar, Saranya Sukumaran, Tripura Panchagnula, Sunjay Murali, Kunal Gandhi, Ambujavalli R, Manickam M, C Vaijayanthi, Krishna… find
  7. Ahrane Mahaganapathy, Sumirtha Karunakaran, Kavitha Navakulan, and Kengatharaiyer Sarveswaran. 2026. Bridging dialectal variation: A phonetic transcription tool for Tamil. In Proceedings of the 13th Workshop on NLP for Similar Languages, Varieties and Dialects, pages 234–241, Rabat, Morocco. Associa… find
  8. Mozilla Foundation. 2019. Mozilla Common Voice. https://commonvoice.mozilla.org. ↗
  9. JILL CR Newbigin. 2019. Evolution of Tamil Language: A Historical Study. Journal of Emerging Technologies and Innovative Research (JETIR), 6(3):115–123. © 2019 JETIR1903O19. find
  10. S. Nishanth, Shruthi Rengarajan, Burugu Rahul, and G. Jyothish Lal. 2025. NSR@LT-EDI-2025: Automatic Speech Recognition in Tamil. In Proceedings of the Fifth Workshop on Language Technology for Equality, Diversity and Inclusion. find
  11. OpenSLR. 2019. OpenSLR: Free Speech and Language Resources. http://www.openslr.org. ↗
  12. Hemant Palivela, Meera Narvekar, David Asirvatham, Shashi Bhushan, Vinay Rishiwal, and Udit Agarwal. 2025. Code-switching asr for low-resource indic languages: A hindi-marathi case study. IEEE Access, 13:9171–9198. find
  13. Kabilan Prasanna and Aryaman Arora. 2024. Irumozhi: Automatically classifying diglossia in Tamil. In Findings of the Association for Computational Linguistics: NAACL 2024, pages 3096– 3103. find
  14. Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2023. Robust speech recognition via large-scale weak supervision. In International Conference on Machine Learning, pages 28492–28518. PMLR.
  15. Francis Rakotomalala, Hasindraibe Randriatsara-fara, Hajalalaina Aimé Richard, and Ravoni-manantsoa Ndaohialy Manda Vy. 2021. Voice User Interface: Literature Review, Challenges and Future Directions. SYSTEM THEORY, CONTROL AND COMPUTING JOURNAL, 1:65–89. find
  16. Kengatharaiyer Sarveswaran, Gihan Dias, and Miriam Butt. 2021. ThamizhiMorph: A morphological parser for the Tamil language. Machine Translation, 35(1):37–70. find
  17. Jubeerathan Thevakumar, Luxshan Thavarasa, Thanikan Sivatheepan, Sajeev Kugarajah, and Uthayasanker Thayasivam. 2025. EmoTa: A Tamil Emotional Speech Dataset. In Proceedings of the First Workshop on Challenges in Processing South Asian Languages (CHiPSAL 2025), pages 193–201, Abu Dhabi, UAE. Interna… find
  18. Preethi Thinakaran, Malarvizhi Muthuramalingam, Anushiya Rachel Gladston, P Vijayalakshmi, Hema A Murthy, T Nagarajan, et al. 2025. SIToBI–A Speech Prosody Annotation Tool for Indian Languages. arXiv preprint arXiv:2502.09661. — [arXiv:2502.09661](https://arxiv.org/abs/2502.09661) find
  19. Naveed Unjum, Stephanie Evert, Kengatharaiyer Sarveswaran, Ruvan Weerasinghe, and Nevidu Jayatilleke. 2026. Llms for low-resource languages: A survey. find
  20. P. Yasmini. 2017. The contrast between jaffna tamil and upcountry tamil: A dialectological study. In The Third International Conference on Linguistics in Sri Lanka, ICLSL 2017, page 143. Department of Linguistics, University of Kelaniya, Sri Lanka. Conference proceedings. find
  21. Zeidan. 2020. Languages by Total Number of Speakers. https://www.britannica.com/topic/languages-by-total-number-of-speakers-2228881. Accessed: 2024-12-29. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/thillainathan-2026-talk-corpus-regionally-diverse-evaluation/