Elamur Research⌕
☰
home / papers / 2026 / Unrequited Emotions: Investigating the Gaps in Motivation and Practice in Speech Emotion Recognition Research

Unrequited Emotions: Investigating the Gaps in Motivation and Practice in Speech Emotion Recognition Research

Paper · 2026 · Taryn Wong, Zeerak Talat, Hanan Aldarmaki et al.

Summary

This survey asks whether speech emotion recognition (SER) research does what it claims to be for. Ethical critiques of emotion AI have mostly targeted facial analysis or generic systems. The authors check SER against a minimal condition: a project's aims, data and emotion concepts should fit together.

They pulled 7,486 candidate papers from Semantic Scholar and narrowed them to 959 from major speech, NLP and ML venues. From these they drew a stratified random sample of 108, and 88 remained after discarding papers not centred on SER. Using inductive coding, they recorded each paper's stated motivations, the emotions it modelled and the datasets it used.

Papers often justify SER with concrete settings such as voice assistants, mental-health care or call centres. Yet experiments rely heavily on a few mostly acted corpora that were never built for those settings. Papers also use only a narrow slice of each corpus's labels. The authors argue this mismatch invites misreading and misuse, and they urge SER work either to adopt data suited to its stated use-cases or to revise its motivations.

Key points

  • Responsive human-computer interaction systems were the dominant motivation, with 42.05% of papers citing unspecified "other HCI systems." Healthcare (18.18%) and call screening (17.05%) followed. 27.27% of papers relied solely on prior SER work for motivation.
  • Six "popular" corpora were each used by at least five papers: IEMOCAP, EMO-DB, RAVDESS, SUSAS, MSP-Improv and RECOLA.
    • Four of these contain acted speech; only SUSAS and RECOLA are spontaneous.
    • 63.64% of papers used at least one of them, and 35.23% used nothing else.
  • IEMOCAP appears in 40.91% of papers and makes up nearly 60% of data use in 2016–2024. Custom-collected datasets (22.73% overall) have become much rarer since 2016.
  • Emotion choice is narrow. Angry (76.14%), neutral, sad and happy (65.91%) are studied about twice as often as any other label. Over 90% of IEMOCAP papers use its categorical labels, but only 2.78% use Dominance.
  • Annotations in these corpora mostly reflect how third parties perceive a speaker's emotion. The authors say this fits poorly with uses such as mental-health support.

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 EdinburghEdinburgh · United Kingdom381 works · 80,747 citations🎓Johns Hopkins UniversityBaltimore · United States248 works · 106,397 citations🎓Mohamed bin Zayed University of Artificial IntelligenceAbu Dhabi · United Arab Emirates138 works · 6,590 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksSocial Sciences · 1,954 worksPsychology · 1,649 worksExperimental and Cognitive Psychology · 930 worksSafety Research · 626 works

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

Cite

Cite as Taryn Wong, Zeerak Talat, Hanan Aldarmaki and Anjalie Field. 2026. Unrequited Emotions: Investigating the Gaps in Motivation and Practice in Speech Emotion Recognition Research. Proceedings of Computational Affective Science (CAS) @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/42e4sre4izeu

@inproceedings{wong2026unrequited,
  title = {{Unrequited Emotions: Investigating the Gaps in Motivation and Practice in Speech Emotion Recognition Research}},
  author = {Taryn Wong, Zeerak Talat, Hanan Aldarmaki and Anjalie Field},
  booktitle = {Proceedings of Computational Affective Science (CAS) @ LREC 2026},
  year = {2026},
  doi = {10.63317/42e4sre4izeu},
  url = {https://doi.org/10.63317/42e4sre4izeu}
}

References

4 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Noam Amir, Ori Kerret, and Dimitry Karlinski. 2001. Classifying emotions in speech: a comparison of methods. In Interspeech. find
  2. Nazanin Andalibi and Justin Buss. 2020. The human in emotion recognition on social media: Attitudes, outcomes, risks. In Proc. of CHI, CHI ’20, page 1–16, New York, NY, USA. Association for Computing Machinery. find
  3. Véronique Aubergé, Nicolas Audibert, and Albert Rilliard. 2003. Why and how to control the authentic emotional speech corpora. In Interspeech. find
  4. Lokesh Kumar Bansal, S. Pavankumar Dubagunta, Malolan Chetlur, Pushpak Jagtap, and Aravind Ganapathiraju. 2023. On the efficacy and noise-robustness of jointly learned speech emotion and automatic speech recognition. In Interspeech. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wong-2026-unrequited-emotions-investigating-gaps-motivation/