Elamur Research⌕
☰
home / papers / 2026 / TamilVoiceLab@DravidianLangTech 2026: Investigating Whisper Tamil Large-v2 for Dialectal Tamil Speech Recognition

TamilVoiceLab@DravidianLangTech 2026: Investigating Whisper Tamil Large-v2 for Dialectal Tamil Speech Recognition

Paper · 2026 · S.B.Priya and Bharathi B

Summary

Speech recognizers trained mostly on standard Tamil often fail on regional varieties, which differ in pronunciation, rhythm, vocabulary and syntax. For the DravidianLangTech 2026 shared task on Tamil Dialect Speech Recognition, the authors build a zero fine-tuning baseline. They apply the pretrained Whisper Tamil Large-v2 model directly, with no extra training or domain adaptation.

The shared-task corpus covers four dialect regions of Tamil Nadu. Its training portion has roughly 9.22 hours of transcribed audio, and a separate test set is unlabeled. Only the test data is used here. The Whisper processor converts 16 kHz audio into log-Mel spectrograms, the model decodes them with beam search, and the output is scored by Word Error Rate.

The system reaches a WER of 0.71. That is clearly worse than the fine-tuned or dialect-trained Tamil systems the authors cite, which report roughly 0.30–0.45. Most errors involve regional vocabulary, colloquial phrasing and dialectal pronunciation, yet many frequent words are still recognized. The study is useful as a reference point for how much dialect-aware adaptation and balanced dialect data Tamil ASR still requires.

Key points

  • Model: Whisper Tamil Large-v2, used off the shelf (no fine-tuning) with log-Mel features and beam search decoding.
  • Data: the Tamil Dialect Speech dataset, released for DravidianLangTech's shared task and described in Bharathi et al. (2026). It covers Northern, Southern, Western and Central dialects, with spontaneous and read speech at 16 kHz.
  • Data splits: the training set has 5,134 recordings (about 9.22 h) and the test set about 2.05 h. The paper gives the number of test files as 579 in some places and 580 in others.
  • Result: WER = 0.71 on the test set. Comparison systems with fine-tuning or LoRA adaptation report about 0.30–0.45.
  • Code is released on GitHub (Prisur2013/TAMILVOICELABS_1991).

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, topics and concepts

📚 Shelves 5The 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 worksSignal Processing · 452 works

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

Cite

Cite as S.B.Priya and Bharathi B. 2026. TamilVoiceLab@DravidianLangTech 2026: Investigating Whisper Tamil Large-v2 for Dialectal Tamil Speech Recognition. Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.dravidianlangtech-1.63

@inproceedings{sbpriya2026tamilvoicelab,
  title = {{TamilVoiceLab@DravidianLangTech 2026: Investigating Whisper Tamil Large-v2 for Dialectal Tamil Speech Recognition}},
  author = {S.B.Priya and Bharathi B},
  booktitle = {Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages},
  year = {2026},
  doi = {10.18653/v1/2026.dravidianlangtech-1.63},
  url = {https://doi.org/10.18653/v1/2026.dravidianlangtech-1.63}
}

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. Priyobroto Acharya, Soham Chaudhuri, Sayan Das, Dipanjan Saha, and Dipankar Das. 2025. Junlp@ lt-edi-2025: Efficient low-rank adaptation of whisper for inclusive tamil speech recognition targeting vulnerable populations. In Proceedings of the 5th Conference on Language, Data and Knowledge: Fifth Wor… find
  2. Maha Tufail Agro, Atharva Kulkarni, Karima Kadaoui, Zeerak Talat, and Hanan Aldarmaki. 2025. Code-switching in end-to-end automatic speech recognition: A systematic literature review. arXiv preprint arXiv:2507.07741. find
  3. Arun Babu, Changhan Wang, Andros Tjandra, Kushal Lakhotia, Qiantong Xu, Naman Goyal, Kritika Singh, Patrick Von Platen, Yatharth Saraf, Juan Pino, and 1 others. 2021. Xls-r: Self-supervised cross-lingual speech representation learning at scale. arXiv preprint arXiv:2111.09296.
  4. Érik Beltrán Lobato. 2025. Multilingual speech-to-speech machine translation. find
  5. B. Bharathi, Bharathi Raja Chakravarthi, Shunmuga Priya Muthusamy Chinnan, S. Saranya, and S. Suhasini. 2026. Findings in Tamil Dialect Speech Recognition and Classification. In Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages. Association for Co… find
  6. B Bharathi, S Saranya, P Vijayalakshmi, and T Nagarajan. 2025. Multi-dialect speech corpus creation for enhancing tamil automatic speech recognition. Circuits, Systems, and Signal Processing, pages 1–19. find
  7. Dawit Ketema Gete, Bedru Yimam Ahmed, Tadesse Destaw Belay, Yohannes Ayana Ejigu, Sukairaj Hafiz Imam, Alemu Belay Tessema, Mohammed Oumer Adem, Tadesse Amare Belay, Robert Geislinger, Umma Aliyu Musa, and 1 others. 2025. Whispering in amharic: Fine-tuning whisper for low-resource language. arXiv pr…
  8. Tahir Javed, Kaushal Bhogale, Abhigyan Raman, Pratyush Kumar, Anoop Kunchukuttan, and Mitesh M Khapra. 2023. Indicsuperb: A speech processing universal performance benchmark for indian languages. In Proceedings of the AAAI conference on artificial intelligence, volume 37, pages 12942–12950. find
  9. Markus Miiller, Sebastian Stiiker, and Alex Waibel. 2018. Multilingual adaptation of rnn based asr systems. In 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 5219–5223. IEEE. find
  10. Gowtham Papala, Aniket Ransing, and Pooja Jain. 2023. Sentiment analysis and speaker diarization in hindi and marathi using using finetuned whisper: sentiment analysis in hindi and marathi. Scalable Computing: Practice and Experience, 24(4):835–846. find
  11. Chandrika Prasad, Veenga Gode Swamy Rao, R China Appala Naidu, and 1 others. 2026. An asr transformer-based model for kannada speech-to-text transcription. Journal of Artificial Intelligence and Technology. find
  12. Alexei Baevski, Yuhao Zhou, Abdelrahman Mohamed, and Michael Auli. 2020. wav2vec 2.0: A framework for self-supervised learning of speech representations. Advances in neural information processing systems, 33:12449–12460.
  13. Xabier de Zuazo, Eva Navas, Ibon Saratxaga, and Inma Hernáez Rioja. 2025. Whisper-lm: Improving asr models with language models for low-resource languages. arXiv preprint arXiv:2503.23542. — [arXiv:2503.23542](https://arxiv.org/abs/2503.23542) find
  14. Yunpeng Liu, Xukui Yang, and Dan Qu. 2024. Exploration of whisper fine-tuning strategies for low-resource asr. EURASIP Journal on Audio, Speech, and Music Processing, 2024(1):29. find
  15. 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.
  16. Frederick Riemenschneider and Anette Frank. 2025. Cross-lingual generalization and compression: from language-specific to shared neurons. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 13470–13491. find
  17. S Saranya, B Bharathi, S Gomathy Dhanya, and Aishwarya Krishnakumar. 2025. Real-time continuous tamil dialect speech recognition and summarization. Circuits, Systems, and Signal Processing, 44(4):2855– 2881. find
  18. Sanket Shah, Kavya Ranjan Saxena, Kancha-rana Manideep Bharadwaj, Sharath Adavanne, and Nagaraj Adiga. 2025. Indicst: Indian multilingual translation corpus for evaluating speech large language models. In 2025 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASS… find
  19. Suwon Shon, Ahmed Ali, and James Glass. 2018. Convolutional neural networks and language embeddings for end-to-end dialect recognition. arXiv preprint arXiv:1803.04567. — [arXiv:1803.04567](https://arxiv.org/abs/1803.04567) find
  20. Xuchen Wei, Yangxin Wu, Yaoyin Zhang, Henglyu Liu, Kehai Chen, Xuefeng Bai, and Min Zhang. 2025. Hitsz’s end-to-end speech translation systems combining sequence-to-sequence auto speech recognition model and indic large language model for iwslt 2025 in indic track. In Proceedings of the 22nd Interna… find
  21. Jiqiao Zhang and Degen Huang. 2025. Speech recognition for low-resource languages using large language models and related-language data. In International Conference on Image, Signal Processing, and Pattern Recognition (ISPP 2025), volume 13664, pages 1264–1269. SPIE. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/priya-2026-tamilvoicelab-dravidianlangtech-2026-investigating-whisper/