KuralHub lists 3 speech emotion recognition datasets for Italian, 2 of them openly available. Each entry below gives the emotion labels, size, recording setup, licence, where to get it and how to cite it. Italian is also part of the benchmark. The best of 11 models reached 80.0% test accuracy (whisper-large).
Italian is one of the 29 languages in the KuralHub benchmark. Each of the 11 frozen speech models was trained with a small classification head on the shared emotions and scored on a held-out test split. Over all 22 runs the mean test accuracy was 50.4% (median 55.5%). The best run reached 80.0% with whisper-large. Compare with other languages.
EMOVO (run ID EMOVO). Scores on the test split, best run highlighted.
DEMoS (Database of Elicited Mood in Speech) is an Italian emotional speech corpus containing 9,365 emotional and 332 neutral samples from 68 native speakers (23 females, 45 males) across seven emotions: anger, sadness, happiness, fear, surprise, disgust, and guilt. Unlike acted speech, DEMoS features emotion elicited through Mood Induction Procedures (MIP). A subset of 1,564 prototypical samples from 59 speakers was selected using self- and external assessment. The dataset has been published in Language Resources and Evaluation.
Citation
@article{parada2020demos,
title={DEMoS: An Italian emotional speech corpus: Elicitation methods, machine learning, and perception},
author={Parada-Cabaleiro, Emilia and Costantini, Giovanni and Batliner, Anton and Schmitt, Maximilian and Schuller, Bj{\"o}rn W},
journal={Language Resources and Evaluation},
volume={54},
number={2},
pages={341--383},
year={2020},
publisher={Springer}
}
EMOVO is the first Italian emotional speech corpus, featuring recordings from 6 actors simulating the Big Six emotions (disgust, fear, anger, joy, surprise, sadness) plus neutrality across 14 sentences. The recordings were captured using professional equipment in the Fondazione Ugo Bordoni laboratories. A validation test with 48 listeners achieved an overall recognition accuracy of 80%, with anger, sadness, and neutrality being the most easily recognized emotions, while joy and disgust were more challenging to identify.
Used in the KuralHub benchmark (run ID EMOVO). See the results above.
Citation
@inproceedings{costantini-etal-2014-emovo,
title = "{EMOVO} Corpus: an {I}talian Emotional Speech Database",
author = "Costantini, Giovanni and
Iaderola, Iacopo and
Paoloni, Andrea and
Todisco, Massimiliano",
editor = "Calzolari, Nicoletta and
Choukri, Khalid and
Declerck, Thierry and
Loftsson, Hrafn and
Maegaard, Bente and
Mariani, Joseph and
Moreno, Asuncion and
Odijk, Jan and
Piperidis, Stelios",
booktitle = "Proceedings of the Ninth International Conference on Language Resources and Evaluation ({LREC}`14)",
month = may,
year = "2014",
address = "Reykjavik, Iceland",
publisher = "European Language Resources Association (ELRA)",
url = "https://aclanthology.org/L14-1478/",
pages = "3501--3504",
abstract = "This article describes the first emotional corpus, named EMOVO, applicable to Italian language,. It is a database built from the voices of up to 6 actors who played 14 sentences simulating 6 emotional states (disgust, fear, anger, joy, surprise, sadness) plus the neutral state. These emotions are the well-known Big Six found in most of the literature related to emotional speech. The recordings were made with professional equipment in the Fondazione Ugo Bordoni laboratories. The paper also describes a subjective validation test of the corpus, based on emotion-discrimination of two sentences carried out by two different groups of 24 listeners. The test was successful because it yielded an overall recognition accuracy of 80{\%}. It is observed that emotions less easy to recognize are joy and disgust, whereas the most easy to detect are anger, sadness and the neutral state."
}
Emozionalmente is an Italian speech emotion dataset containing 6,902 labeled audio samples from 431 amateur actors expressing the Big Six emotions (anger, disgust, fear, joy, sadness, surprise) plus neutrality. The recordings are 16-bit mono .wav files at 16 kHz, averaging 3.81 seconds. Emotional labels represent the actors' intended emotions, validated by 829 human annotators with 66% accuracy. The dataset includes audio files and three CSVs detailing actor demographics, evaluator data, and emotion labels.
Used in the KuralHub benchmark (run ID Emozionalmente). See the results above.
Citation
@ARTICLE{10879457,
author={Catania, Fabio and Wilke, Jordan W. and Garzotto, Franca},
journal={IEEE Transactions on Audio, Speech and Language Processing},
title={Emozionalmente: A Crowdsourced Corpus of Simulated Emotional Speech in Italian},
year={2025},
volume={},
number={},
pages={1-14},
keywords={Speech recognition;Crowdsourcing;Emotion recognition;Audio recording;Data collection;Computational modeling;Accuracy;Training;Electronic mail;Digital audio broadcasting;Affective computing;Speech emotion recognition;Corpus;Dataset;Italian;Crowdsourcing},
doi={10.1109/TASLPRO.2025.3540662}}
Know another Italian speech emotion dataset, or a newer version of one listed here? Suggest it or open a pull request.
This page is generated from datasets/Italian/ in the KuralHub repository.
If you use a dataset, cite its authors. If KuralHub helped you find it, please also cite the KuralHub paper.