AVA-Speech
收藏资源简介:
AVA-Speech是由谷歌公司创建的一个包含190部电影视频的数据集,旨在为语音活动检测提供一个公开可用的基准。该数据集详细标注了三种语音活动条件:清晰语音、语音与音乐共存、语音与噪音共存,以分析模型在不同背景噪声下的性能。创建过程中,数据集从YouTube视频中选取,并通过人工标注确保每个时刻都有对应的标签。AVA-Speech适用于语音识别、语言识别和说话人分割等应用,旨在解决现有数据集不易获取和比较的问题。
AVA-Speech is a dataset containing 190 movie videos created by Google, which serves as a publicly available benchmark for speech activity detection. It provides detailed annotations for three speech activity scenarios: clear speech, speech coexisting with music, and speech coexisting with background noise, to facilitate the analysis of model performance under different background noise conditions. During the dataset development process, the samples were selected from YouTube videos, and manual annotations were carried out to ensure that each temporal segment has a corresponding label. AVA-Speech is suitable for applications including speech recognition, language recognition and speaker segmentation, and it aims to solve the problems that existing datasets are not easily accessible and hard to compare.




