Word-level Persian Lipreading Dataset

本研究介绍了名为‘Word-level Persian Lipreading Dataset’的数据集，由伊朗科技大学计算机工程学院创建。该数据集包含244,000个视频，涉及约1,800名演讲者，视频来自波斯语视频流网站Aparat。数据集涵盖多种光照条件和姿态，包含205小时的视频内容，主要用于波斯语的单词级唇读研究。创建过程中，研究者采用了包括视频选择、面部跟踪提取、主动说话者检测等自动化流程。该数据集旨在解决波斯语唇读识别的挑战，特别是在音频信号受干扰的环境中提高识别准确性。

This study introduces a dataset named 'Word-level Persian Lipreading Dataset', which was developed by the School of Computer Engineering, Iran University of Science and Technology. The dataset consists of 244,000 videos involving approximately 1,800 speakers, sourced from the Persian video streaming platform Aparat. It covers diverse lighting conditions and poses, with a total of 205 hours of video content, and is primarily designed for word-level Persian lipreading research. During the dataset construction, the researchers adopted automated workflows including video selection, facial tracking and extraction, active speaker detection and other related processes. This dataset aims to address the challenges of Persian lipreading recognition, especially to improve recognition accuracy in environments where audio signals are disturbed.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于波斯语单词级唇读研究的大规模视频数据集，由伊朗科技大学创建，包含244,000个视频，涉及约1,800名演讲者，总时长达205小时，视频来源于波斯语网站Aparat，涵盖多种光照和姿态条件。它通过自动化流程（如面部跟踪和说话者检测）构建，旨在解决波斯语唇读识别的挑战，特别是在音频受干扰环境中提高识别准确性。