KE-Team/SemanticVAD-Dataset

SemanticVAD对话状态检测数据集是一个用于全双工人机语音交互系统的语义级语音活动检测任务的训练与测试数据集。包含15,000条中文和英文的训练样本，以及4,400条中文和英文的测试样本。数据集标注质量经过大模型验证优化，支持人机对话中的发言打断、附和、未完和完成四种状态检测。

The SemanticVAD Dialogue State Detection Dataset is a training and testing dataset for semantic voice activity detection in full-duplex human-machine voice interaction systems. It includes 15,000 training samples in both Chinese and English, and 4,400 testing samples in both Chinese and English. The dataset has been verified and optimized by large-scale models, supporting detection of four states in human-machine dialogue: interruption, agreement, incomplete, and complete.