TED Emotion

该数据集基于野生TED演讲视频，用于情感手势合成，包含了78,734个带有情感标签的手势片段。该数据集被划分为训练集（66,679个片段）、验证集（6,258个片段）和测试集（5,797个片段），并使用预训练的基于音频的情感分类器对手势片段进行情感标注。规模上，该数据集共有78,734个手势片段，其任务是对情感手势进行标注和合成。

This dataset is derived from uncurated TED talk videos and is designed for emotional gesture synthesis. It encompasses 78,734 gesture segments annotated with emotional labels. The dataset is divided into three splits: the training set (66,679 segments), validation set (6,258 segments), and test set (5,797 segments). Emotional annotations for all gesture segments were generated using a pre-trained audio-based emotion classifier. In total, this dataset contains 78,734 gesture segments, with its target tasks being emotional gesture annotation and synthesis.