YODAS

YODAS是由卡内基梅隆大学创建的大型多语种音频和语音数据集，包含超过50万小时的语音数据，涵盖100多种语言。数据集分为手动标注、自动标注和未标注三个子集，分别用于监督学习和自监督学习。YODAS的创建过程涉及从YouTube收集数据，并通过特定的数据收集架构进行处理，以确保数据的质量和多样性。该数据集主要应用于语音识别领域，旨在解决大规模多语种语音数据的可用性问题。

YODAS is a large-scale multilingual audio and speech dataset created by Carnegie Mellon University. It contains over 500,000 hours of speech data covering more than 100 languages. The dataset is divided into three subsets: manually annotated, automatically annotated, and unannotated, which are respectively used for supervised learning and self-supervised learning. The creation of YODAS involves collecting data from YouTube and processing it through a dedicated data collection architecture to ensure data quality and diversity. This dataset is mainly applied in the field of speech recognition, aiming to solve the availability problem of large-scale multilingual speech data.