数据堂—155小时唇形同步视频数据_多模态

249人参与录制语音以及相匹配的唇语视频，多设备同步录制，通过脉冲信号进行精准对齐，准确性高。可用于语音图像领域的多模态学习算法研究。经多家AI公司验证：有助于模型面对真实世界的多样性时能够表现出色。我们严格遵循数据保护法规和隐私规定，确保数据采集、存储和使用的过程中维护用户的隐私和合法权益，所有数据均遵循GDPR, CCPA, PIPL

A total of 249 participants volunteered to record speech and matching lip-reading videos. All recordings were conducted with synchronized multi-device setups and precisely aligned via pulse signals to ensure high accuracy. This dataset is suitable for research on multimodal learning algorithms in the speech and image domains. Verified by multiple AI companies, it helps models achieve excellent performance when facing the diversity of real-world scenarios. We strictly comply with data protection regulations and privacy rules, ensuring that user privacy and legitimate rights and interests are safeguarded throughout the entire process of data collection, storage and utilization. All datasets adhere to the requirements of GDPR, CCPA and PIPL.

搜集汇总

背景与挑战

背景概述

该数据集包含155小时的唇形同步多模态视频数据，专为语音和图像领域的多模态学习算法研究设计。数据由249名中国参与者录制，涵盖六个视频角度和两个音频距离，提供高精度的语音与唇部运动同步信息。