Elamur Research⌕
☰
home / papers / 2025 / EmoTale: An Enacted Speech-emotion Dataset in Danish

EmoTale: An Enacted Speech-emotion Dataset in Danish

Paper · 2025 · Maja J. Hjuler, Harald V. Skat-R\ordam, Line H. Clemmensen et al.

Summary

Large emotional speech corpora exist for widely spoken languages, but Danish has had almost nothing beyond DES, a small 1997 database that was not built for speech emotion recognition (SER). The authors introduce EmoTale, a corpus of acted emotional speech in Danish, with an optional English part. Speakers with acting experience, including children, read five neutral, context-poor sentences adapted from Emo-DB, each in five emotions. Three annotators labelled every clip with a category and with arousal, valence and dominance scores.

To check that the recordings carry a usable emotional signal, the authors take two steps. First, they compare the human dimensional ratings with the outputs of a pretrained wav2vec2 model. Second, they train SER classifiers on EmoTale and on reference corpora in German, Urdu, Danish and Greek. Self-supervised embeddings outperform openSMILE hand-crafted features. The best model reaches 64.1% UAR under leave-one-speaker-out evaluation, which is close to the results on DES. The release gives Danish a test bed for checking SER systems, including on children, before deployment.

Key points

  • EmoTale has 450 Danish and 350 English recordings from 18 participants (12 female, 6 male), aged 9–39 with a mean of 22.8. Clips were recorded at 48 kHz WAV with RØDE Wireless Go microphones.
  • There are five sentences translated from Emo-DB and five emotions: Neutral, Anger, Sadness, Happiness and Boredom. Filenames encode language, speaker, emotion and sentence, for example DK_004_A_5.wav.
  • Agreement on categories, measured with Cohen's κ, was 0.71 between the two annotators. Agreement with the intended emotion was 0.75 and 0.85. Agreement on dimensions, measured with CCC, was 0.72 for arousal, 0.75 for valence and 0.57 for dominance.
  • The validation compares human labels with w2v2-FT-dim, a wav2vec2 model fine-tuned on MSP-Podcast. The reference datasets are Emo-DB, Urdu, DES and AESDD.
  • SSL embeddings beat openSMILE features, and the best leave-one-speaker-out UAR is 64.1%, comparable to performance on DES.

Interesting ideas

  • I1. The carrier sentences are deliberately neutral and nearly context-free, so the label can only come from how they are delivered, not from the words
  • I2. Annotators independently guess the intended category, and their κ against the intended label measures how recoverable the intended signal is
  • I3. An off-the-shelf pretrained model serves as an extra rater, and its agreement with humans on the new corpus is compared against the same agreement on established corpora
  • I4. Leave-one-speaker-out cross-validation stops a model from scoring well just by recognising individual speakers

For Mova

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

  • I1 (frames): build irony and speech-act test items where the same neutral sentence sits under different preceding contexts. This forces prag to rely on context rather than lexical cues.
  • I2 (learn): when an LLM teacher generates examples with a target label, have independent witnesses re-label them. Then keep only batches whose agreement with the intended label passes a threshold.
  • I3 (tree): when converting or adding a treebank, run the witness parsers on it and on a standard UD treebank. An unusually low score on the new data then flags conversion or annotation problems.
  • I4 (answer): evaluate the fable and tale QA readers leave-one-text-out, so memorised story-specific patterns cannot inflate the accuracy.

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

🔬Centre National de la Recherche ScientifiqueParis · France169 works · 62,488 citations🎓Université Grenoble AlpesSaint-Martin-d'Hères · France24 works · 8,259 citations🔬Laboratoire d'Informatique de GrenobleGrenoble · France24 works · 689 citations🎓Technical University of DenmarkKongens Lyngby · Denmark4 works · 698 citations🎓Institut polytechnique de GrenobleGrenoble · France11 works · 207 citations

 Fields, topics and concepts

📚 Shelves 7The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksPsychology · 1,649 worksExperimental and Cognitive Psychology · 930 worksSignal Processing · 493 works

🧭 Topics 2Research subjects and languages the work is about.Sentiment & emotion · 1,035 worksSpeech & spoken language · 519 works

Cite

Cite as Maja J. Hjuler, Harald V. Skat-R\ordam, Line H. Clemmensen and Sneha Das. 2025. EmoTale: An Enacted Speech-emotion Dataset in Danish. arXiv:2508.14548.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2508.14548 · arXiv 2508.14548

@misc{hjuler2025emotale,
  title = {{EmoTale: An Enacted Speech-emotion Dataset in Danish}},
  author = {Maja J. Hjuler, Harald V. Skat-Rordam, Line H. Clemmensen and Sneha Das},
  year = {2025},
  doi = {10.48550/arxiv.2508.14548},
  eprint = {2508.14548},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2508.14548}
}

References

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

  1. K. A. Lindquist, “The role of language in emotion: existing evidence and future directions,” Current Opinion in Psychology, vol. 17, pp. 135–139, 2017. ↗
  2. G. Fairbanks and W. Pronovost, “Vocal pitch during simulated emotion,” Science, vol. 88, no. 2286, pp. 382–383, 1938. ↗
  3. W. F. Soskin and P. E. Kauffman, “Judgment of emotion in word-free voice samples.” Journal of Communication, 1961. ↗
  4. F. Dellaert, T. Polzin, and A. Waibel, “Recognizing emotion in speech,” in Proceeding of Fourth International Conference on Spoken Language Processing. ICSLP’96, vol. 3. IEEE, 1996, pp. 1970–1973. find
  5. B. W. Schuller, “Speech emotion recognition: Two decades in a nutshell, benchmarks, and ongoing trends,” Communications of the ACM, vol. 61, no. 5, pp. 90–99, 2018. find
  6. E. B. Kang, “On the praxes and politics of ai speech emotion recognition,” in Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency, 2023, pp. 455–466. find
  7. G. N. Yannakakis, R. Cowie, and C. Busso, “The ordinal nature of emotions,” in 2017 Seventh International Conference on Affective Computing and Intelligent Interaction (ACII). IEEE, 2017, pp. 248– 255. find
  8. P. Ekman, “Are there basic emotions?” 1992. ↗
  9. C. Busso, M. Bulut, C.-C. Lee, A. Kazemzadeh, E. Mower, S. Kim, J. N. Chang, S. Lee, and S. S. Narayanan, “Iemocap: interactive emotional dyadic motion capture database,” Language Resources and Evaluation, vol. 42, no. 4, pp. 335–359, 2008. ↗
  10. R. Lotfian and C. Busso, “Building naturalistic emotionally balanced speech corpus by retrieving emotional speech from existing podcast recordings,” Ieee Transactions on Affective Computing, vol. 10, no. 4, pp. 471–483, 2019. find
  11. C. Busso, S. Parthasarathy, A. Burmania, M. AbdelWahab, N. Sadoughi, and E. M. Provost, “Msp-improv: An acted corpus of dyadic interactions to study emotion perception,” IEEE Transactions on Affective Computing, vol. 8, no. 1, pp. 67–80, 2016. ↗
  12. I. Engberg, A. Hansen, O. Andersen, and P. Dalsgaard, “Design recording and verification of a danish emotional speech database,” in EU-ROSPEECH’97 : 5th European Conference on Speech Communication and Technology, Patras, Rhodes, Greece, 22-25 September 1997, 1997, pp. Vol. 4, pp. 1695–1698, design Recording and Verification of a Danish Emotional Speech Database; Conference date: 19-05-2010. find
  13. J. Wagner, A. Triantafyllopoulos, H. Wierstorf, M. Schmitt, F. Burkhardt, F. Eyben, and B. W. Schuller, “Dawn of the transformer era in speech emotion recognition: Closing the valence gap,” Ieee Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 9, pp. 10 745– 10 759, 2023. ↗
  14. Y. Wang, A. Boumadane, and A. Heba, “A fine-tuned wav2vec 2.0/hubert benchmark for speech emotion recognition, speaker verification and spoken language understanding,” p. 7, 2022. ↗
  15. L. Pepino, P. Riera, and L. Ferrer, “Emotion recognition from speech using wav2vec 2.0 embeddings,” p. 5, 2021. find
  16. Z. Zhang, X. Zhang, M. Guo, W. Q. Zhang, K. Li, and Y. Huang, “A multilingual framework based on pre-training model for speech emotion recognition,” 2021 Asia-pacific Signal and Information Processing Association Annual Summit and Conference, Apsipa Asc 2021 - Proceedings, pp. 750–755, 2021. find
  17. S.-w. Yang, P.-H. Chi, Y.-S. Chuang, C.-I. J. Lai, K. Lakhotia, Y. Y. Lin, A. T. Liu, J. Shi, X. Chang, G.-T. Lin, T.-H. Huang, W.-C. Tseng, K.-t. Lee, D.-R. Liu, Z. Huang, S. Dong, S.-W. Li, S. Watanabe, A. Mohamed, and H.-y. Lee, “Superb: Speech processing universal performance benchmark,” 2021. ↗
  18. F. Burkhardt, A. Paeschke, M. Rolfes, W. Sendlmeier, and B. Weiss, “A database of german emotional speech,” vol. 5, 09 2005, pp. 1517–1520. find
  19. “Ethical approval application to the IRB for DanskEmoTale: a pilot study,” https://github.com/DTUComputeStatisticsAndDataAnalysis/Analysis-of-emotions-using-physiological-signals-a-pilot-study/blob/main/Analysis plan modifed.pdf, Sept 2022. ↗
  20. T. Gebru, J. Morgenstern, B. Vecchione, J. W. Vaughan, H. Wallach, H. D. Iii, and K. Crawford, “Datasheets for datasets,” Communications of the ACM, vol. 64, no. 12, pp. 86–92, 2021. ↗
  21. S. Das, N. N. Lonfeldt, N. L. Lund, A. K. Pagsberg, and L. K. H. Clemmensen, “Zero-shot cross-lingual speech emotion recognition: A study of loss functions and feature importance,” Proceedings of 2nd Symposium on Security and Privacy in Speech Communication, 2022. find
  22. S. Das, N. N. Lønfeldt, A. K. Pagsberg, L. Clemmensen et al., “Speech detection for child-clinician conversations in danish for low-resource in-the-wild conditions: a case study,” arXiv preprint arXiv:2204.11550, 2022. ↗
  23. J. Cohen, “A coefficient of agreement for nominal scales,” Educational and Psychological Measurement, vol. 20, no. 1, pp. 37–46, 1960. [Online]. Available: https://doi.org/10.1177/001316446002000104 ↗
  24. I. Siegert, R. Böck, and A. Wendemuth, “Inter-rater reliability for emotion annotation in human-computer interaction – comparison and methodological improvements,” Journal of Multimodal User Interfaces, vol. 8, pp. 17–28, 01 2014. ↗
  25. I. Lawrence and K. Lin, “A concordance correlation coefficient to evaluate reproducibility,” Biometrics, pp. 255–268, 1989. ↗
  26. S. Latif, A. Qayyum, M. Usman, and J. Qadir, “Cross lingual speech emotion recognition: Urdu vs. western languages,” in 2018 International conference on frontiers of information technology (FIT). IEEE, 2018, pp. 88–93. ↗
  27. N. Vryzas, R. Kotsakis, A. Liatsou, C. A. Dimoulas, and G. Kalliris, “Speech emotion recognition for performance interaction,” Journal of the Audio Engineering Society, vol. 66, no. 6, pp. 457–467, 2018. find
  28. S. Das, N. L. Lund, N. N. Lønfeldt, A. K. Pagsberg, and L. H. Clemmensen, “Continuous metric learning for transferable speech emotion recognition and embedding across low-resource languages,” in Proceedings of the Northern Lights Deep Learning Workshop, vol. 3, 2022. find
  29. F. Eyben, K. R. Scherer, B. W. Schuller, J. Sundberg, E. André, C. Busso, L. Y. Devillers, J. Epps, P. Laukka, S. S. Narayanan et al., “The geneva minimalistic acoustic parameter set (gemaps) for voice research and affective computing,” IEEE transactions on affective computing, vol. 7, no. 2, pp. 190–202, 2015. ↗
  30. B. Schuller, S. Steidl, A. Batliner, F. Burkhardt, L. Devillers, C. Müller, and S. S. Narayanan, “The interspeech 2010 paralinguistic challenge,” Interspeech 2010, 2010. find
  31. F. Eyben, M. Wöllmer, and B. Schuller, “Opensmile: the munich versatile and fast open-source audio feature extractor,” in Proceedings of the 18th ACM international conference on Multimedia, 2010, pp. 1459–1462. find
  32. A. Baevski, H. Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representations,” 2020.
  33. S. R. Livingstone and F. A. Russo, “The Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS),” Apr. 2018. [Online]. Available: https://doi.org/10.5281/zenodo.1188976 ↗
  34. J. Wagner, A. Triantafyllopoulos, H. Wierstorf, M. Schmitt, F. Burkhardt, F. Eyben, and B. W. Schuller, “Model for dimensional speech emotion recognition based on wav2vec 2.0,” feb 2022. [Online]. Available: https://doi.org/10.5281/zenodo.6221127 ↗
  35. B. Schuller, S. Steidl, A. Batliner, J. Hirschberg, J. K. Burgoon, A. Baird, A. Elkins, Y. Zhang, E. Coutinho, and K. Evanini, “The INTERSPEECH 2016 Computational Paralinguistics Challenge: Deception, Sincerity & Native Language,” in Proc. Interspeech 2016, 2016, pp. 2001–2005. find
  36. P. B. Brockhoff, J. K. Møller, E. W. Andersen, P. Bacher, and L. E. Christiansen, “Introduction to statistics at dtu,” 2018. find
  37. S. Das, N. N. Lønfeldt, A. K. Pagsberg, and L. H. Clemmensen, “Towards transferable speech emotion representation: on loss functions for cross-lingual latent representations,” in ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2022, pp. 6452–6456. ↗
  38. T. Gebru, J. Morgenstern, B. Vecchione, J. W. Vaughan, H. Wallach, H. D. III, and K. Crawford, “Datasheets for datasets,” 2021. [Online]. Available: https://arxiv.org/abs/1803.09010 ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/hjuler-2025-emotale-enacted-speech-emotion-dataset/