KuralHub lists 2 speech emotion recognition datasets for Turkish, 1 of them openly available. Each entry below gives the emotion labels, size, recording setup, licence, where to get it and how to cite it. Turkish is also part of the benchmark. The best of 11 models reached 84.0% test accuracy (whisper-small).
Turkish 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 11 runs the mean test accuracy was 54.9% (median 48.0%). The best run reached 84.0% with whisper-small. Compare with other languages.
TurEV-DB (run ID TurEV-DB). Scores on the test split, best run highlighted.
The dataset consists of 5304 speech signals from 55 Turkish movies, with their textual contents extracted. These speech signals are labeled both categorically (happy, surprised, sad, angry, fear, neutral, and other) and in a 3-dimensional emotional space (valence, activation, and dominance). It is designed for emotion recognition and acoustic analysis of Turkish emotional speech.
Citation
@INPROCEEDINGS{5929860,
author={Oflazoglu, Çağlar and Yildirim, Serdar},
booktitle={2011 IEEE 19th Signal Processing and Communications Applications Conference (SIU)},
title={Turkish emotional speech database},
year={2011},
volume={},
number={},
pages={1153-1156},
keywords={Speech;Databases;Conferences;Psychology;Signal processing;TV;Emotion recognition},
doi={10.1109/SIU.2011.5929860}}
The Turkish Emotion-Voice Database (TurEV-DB) is a corpus of over 1735 tokens, consisting of 82 words uttered by human subjects across four emotions: angry, calm, happy, and sad. The speech data were produced by amateur actors, processed through a denoising procedure, and checked by assessors. The database includes the speech corpus, spectrograms, extracted prosodic features, prosodic data graphs, and model activations. It also supports machine learning experiments using CNN and SVM models, with frequency-filtered speech signals across three frequency ranges (0-8000 Hz, 0-5000 Hz, 500-8000 Hz) for improved emotion detection.
Used in the KuralHub benchmark (run ID TurEV-DB). See the results above.
Citation
@inproceedings{canpolat-etal-2020-turkish,
title = "{T}urkish Emotion Voice Database ({T}ur{EV}-{DB})",
author = "Canpolat, Salih Firat and
Ormano{\u{g}}lu, Zuhal and
Zeyrek, Deniz",
editor = "Beermann, Dorothee and
Besacier, Laurent and
Sakti, Sakriani and
Soria, Claudia",
booktitle = "Proceedings of the 1st Joint Workshop on Spoken Language Technologies for Under-resourced languages (SLTU) and Collaboration and Computing for Under-Resourced Languages (CCURL)",
month = may,
year = "2020",
address = "Marseille, France",
publisher = "European Language Resources association",
url = "https://aclanthology.org/2020.sltu-1.52/",
pages = "368--375",
language = "eng",
ISBN = "979-10-95546-35-1",
abstract = "We introduce the Turkish Emotion-Voice Database (TurEV-DB) which involves a corpus of over 1700 tokens based on 82 words uttered by human subjects in four different emotions (\textit{angry, calm, happy, sad}). Three machine learning experiments are run on the corpus data to classify the emotions using a convolutional neural network (CNN) model and a support vector machine (SVM) model. We report the performance of the machine learning models, and for evaluation, compare machine learning results with the judgements of humans."
}
Know another Turkish speech emotion dataset, or a newer version of one listed here? Suggest it or open a pull request.
This page is generated from datasets/Turkish/ in the KuralHub repository.
If you use a dataset, cite its authors. If KuralHub helped you find it, please also cite the KuralHub paper.