AVA

该数据集名为AVA，包含了用于主动说话者检测的视频片段，其中也包括了配音电影。然而，这些配音电影由于存在不同步的说话片段，可能会导致训练和评估出现误导。特别是在AVA数据集中，这些配音电影对于训练和评估来说存在问题，因为它们包含了被标记为说话但实际上并未同步的说话片段。该任务的目的是进行主动说话者检测。This dataset, named AVA, contains video clips intended for active speaker detection (ASD) tasks, including dubbed films. However, these dubbed movies may introduce misleading outcomes during model training and evaluation, as they possess out-of-sync speech segments. Specifically, the dubbed movies included in the AVA dataset are problematic for both training and evaluation processes, since they contain speech segments that are annotated as corresponding to speaking behaviors but are actually not synchronized. The target task of this dataset is active speaker detection.

搜集汇总

背景与挑战

背景概述

AVA数据集用于主动说话者检测，包含视频片段，其中涉及配音电影，但这些配音电影存在说话片段不同步的问题，可能导致训练和评估时的误导。