Earnings-22

Earnings-22数据集是由Rev.com开发的，旨在为自动语音识别（ASR）系统提供一个包含多种口音的实际音频基准。该数据集包含来自全球27个国家的125个英语财报电话录音，总时长119小时，涵盖7种地区口音。数据集的创建过程涉及从多个来源收集录音，并通过专业的人工转录平台进行高质量转录。Earnings-22数据集主要用于学术和工业研究，以解决ASR系统在处理真实世界口音音频时的性能问题，推动更公平和有效的语音技术发展。

The Earnings-22 dataset was developed by Rev.com, designed to serve as a real-world audio benchmark with diverse accents for automatic speech recognition (ASR) systems. This dataset comprises 125 English earnings call recordings from 27 countries worldwide, with a total duration of 119 hours and coverage of 7 regional accents. The creation of the Earnings-22 dataset involved collecting recordings from multiple sources and performing high-quality transcriptions via professional manual transcription platforms. The Earnings-22 dataset is primarily utilized for academic and industrial research, aiming to resolve the performance issues of ASR systems when processing real-world accented audio and promote the development of more equitable and effective speech technologies.