HUME-VB

该数据集是一个大规模的情感非语言发声（即发声爆发）数据集，包含了来自1,702名发言者的59,299个发声样本。这些样本在参与者家中录制，模仿了种子情感的例子，并对10种不同的表达情绪进行了标注，每种情绪的强度评分范围在1到100之间。这是一个大型数据集，拥有59,299个发声样本，适用于多标签回归、二维情感建模、文化方面探索以及发声爆发分类等任务。This is a large-scale dataset of emotional non-verbal vocalizations (i.e., vocal bursts) containing 59,299 samples from 1,702 speakers. All samples were recorded in participants' homes, with speakers imitating examples of seed emotions during data collection. Each sample was annotated with 10 distinct expressed emotions, and the intensity score for each emotion ranges from 1 to 100. This dataset, with its 59,299 vocal samples, is suitable for tasks including multi-label regression, two-dimensional emotion modeling, cultural exploration, and vocal burst classification.

搜集汇总

背景与挑战

背景概述

HUME-VB数据集是一个用于声乐爆发情感分析的竞赛数据集，包含四个任务：高维情感预测（10种情感）、二维情感预测（唤醒度和效价）、跨文化情感预测（针对美国、中国、委内瑞拉和南非）以及爆发类型分类（8个类别）。数据集提供了基线性能指标（如高维任务CCC为0.5686）和标准化的提交格式，旨在推动多输出回归和分类模型在跨文化情感计算中的应用。