LRW

该数据集名为LRW，是一个大规模的英文音视频词级别数据集，包含500个单词，每个单词最多有1000个训练视频。数据集经过预处理，视频围绕嘴唇区域裁剪，调整为灰度图像，音频则转换成梅尔频谱图。该数据集的规模属于大型，其任务是对唇语进行识别。

This dataset, named LRW, is a large-scale English audio-visual word-level dataset. It contains 500 distinct words, with up to 1000 training videos per word. The dataset has been preprocessed: videos are cropped around the lip region, converted to grayscale images, and audio is transformed into mel-spectrograms. As a large-scale dataset, its core task is lip reading recognition.