Voxlect
收藏资源简介:
Voxlect是一个针对全球方言和地区语言建模的语音基础模型基准数据集。该数据集由南加州大学的研究团队创建,包含超过200万个训练语音样本,来自30个公开可用的语音语料库,涵盖了英语、阿拉伯语、普通话和粤语、藏语、印地语、泰语、西班牙语、法语、德语、巴西葡萄牙语和意大利语等多种语言的方言和地区语言变体。Voxlect旨在解决自动语音识别(ASR)系统在不同方言和地区语言变体之间的性能差异问题,通过建模和识别不同的方言,可以更好地理解当前语音技术的局限性,并推动更可靠、更鲁棒的语音技术的发展。
Voxlect is a foundational speech model benchmark dataset targeting global dialects and regional languages. Developed by a research team at the University of Southern California, it contains over 2 million training speech samples sourced from 30 publicly available speech corpora, covering dialect and regional language variants of multiple languages including English, Arabic, Mandarin, Cantonese, Tibetan, Hindi, Thai, Spanish, French, German, Brazilian Portuguese, and Italian. Voxlect aims to address the performance disparity of automatic speech recognition (ASR) systems across different dialects and regional language variants. By modeling and recognizing diverse dialects, it enables a better understanding of the limitations of current speech technologies and promotes the development of more reliable and robust speech technologies.



