OWLS Multilingual Speech Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://huggingface.co/collections/espnet/owls-scaling-laws-for-speech-recognition-and-translation-67ab7f991c194065f057ce8d
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了高达36万小时的公共演讲数据，跨越150种语言，用于训练和评估多语言语音识别及翻译模型。此外，该数据集还包括了针对不同模型大小（从25亿到1800亿参数）的多语言自动语音识别和语音到文本翻译的评估，并特别为资源较少的语言设定了基准测试。其规模达到36万小时，任务涵盖了多语言语音识别和翻译。
This dataset contains up to 360,000 hours of public speech data spanning 150 languages, designed for training and evaluating multilingual speech recognition and translation models. It also includes evaluations of multilingual automatic speech recognition and speech-to-text translation across model sizes ranging from 2.5 billion to 180 billion parameters, with specially established benchmark tests for low-resource languages. Boasting a corpus size of 360,000 hours, the dataset covers tasks including multilingual speech recognition and translation.
提供机构：
Public speech data sources
