Embedded Systems Laboratory, Ecole Polytechnique Fédérale de Lausanne (EPFL), Vaud, Switzerland.
Instituto de Estudios de Género, Universidad Carlos III de Madrid, Calle Madrid, 126, 28903, Getafe (Madrid), España.
Sci Data. 2024 Oct 30;11(1):1182. doi: 10.1038/s41597-024-04002-8.
WEMAC is a unique open multi-modal dataset that comprises physiological, speech, and self-reported emotional data records of 100 women, targeting Gender-based Violence detection. Emotions were elicited through visualizing a validated video set using an immersive virtual reality headset. The physiological signals captured during the experiment include blood volume pulse, galvanic skin response, and skin temperature. The speech was acquired right after the stimuli visualization to capture the final traces of the perceived emotion. Subjects were asked to annotate among 12 categorical emotions, several dimensional emotions with a modified version of the Self-Assessment Manikin, and liking and familiarity labels. The technical validation proves that all the targeted categorical emotions show a strong statistically significant positive correlation with their corresponding reported ones. That means that the videos elicit the desired emotions in the users in most cases. Specifically, a negative correlation is found when comparing fear and not-fear emotions, indicating that this is a well-portrayed emotional dimension, a specific, though not exclusive, purpose of WEMAC towards detecting gender violence.
WEMAC 是一个独特的开放多模态数据集，包含 100 名女性的生理、语音和自我报告的情感数据记录，旨在进行基于性别的暴力检测。通过使用沉浸式虚拟现实耳机可视化经过验证的视频集来引发情感。实验过程中采集的生理信号包括血流量脉冲、皮肤电反应和皮肤温度。在刺激可视化后立即获取语音，以捕捉感知情感的最终痕迹。要求受试者在 12 个类别情绪、经过修改的自我评估情绪和喜欢与熟悉度标签之间进行标注。技术验证表明，所有目标类别情绪都与其对应的报告情绪呈强烈的统计上显著正相关。这意味着在大多数情况下，视频会在用户中引起预期的情绪。特别是，在比较恐惧和不恐惧情绪时发现了负相关，这表明这是一个很好地描绘的情感维度，这是 WEMAC 检测性别暴力的一个特定但非排他性的目的。
