REWIND

REWIND数据集是由代尔夫特理工大学的研究团队创建，专注于从多模态身体运动信号中进行隐私保护的说话状态分割。该数据集包含33名参与者在专业网络活动中的高质量个人语音录音，提供了视频、姿态和可穿戴加速度等多种模态的数据。创建过程中，通过自动化的姿态注释和语音处理算法自动获取说话活动标签，实现了高时间分辨率的说话状态信号提取。REWIND数据集的应用领域广泛，旨在解决社交互动中说话状态的自动检测问题，特别是在无法直接获取音频的复杂环境中，为社交信号处理和计算社交科学提供了新的研究方向。The REWIND dataset was created by a research team from Delft University of Technology, focusing on privacy-preserving speech state segmentation from multimodal bodily movement signals. This dataset contains high-quality personal speech recordings from 33 participants during professional networking events, and provides multimodal data including video, pose, and wearable acceleration measurements. During its construction, speech activity labels were automatically acquired via automated pose annotation and speech processing algorithms, enabling high-temporal-resolution extraction of speech state signals. The REWIND dataset has wide-ranging applications, aiming to address the automatic detection of speech state in social interactions, especially in complex environments where direct audio access is unavailable, and provides new research directions for social signal processing and computational social science.

搜集汇总

背景与挑战

背景概述

REWIND数据集由代尔夫特理工大学创建，专注于从多模态身体运动信号中进行隐私保护的说话状态分割，包含33名参与者的视频、姿态和可穿戴加速度数据，通过自动化算法实现高分辨率标签提取，应用于社交互动中的说话状态自动检测，为复杂环境下的社交信号处理提供支持。