AfriSpeech-200

AfriSpeech-200是由Masakhane NLP等多个研究机构合作创建的开放源代码非洲口音英语语音数据集，旨在推动非洲口音在临床和通用领域的自动语音识别（ASR）研究。该数据集包含67,577个语音-转录对，涵盖13个国家的120种非洲本土口音，总时长达到200.7小时。创建过程中，研究团队通过网络爬虫和模板技术，从多个非洲新闻网站和医疗相关网站收集文本，增加数据集的多样性和相关性。AfriSpeech-200的应用领域广泛，旨在解决非洲口音在ASR系统中的代表性不足问题，尤其是在医疗记录等关键领域的应用，以提高非洲医疗服务的效率和质量。

AfriSpeech-200 is an open-source accented English speech dataset for African English accents, co-created by Masakhane NLP and multiple research institutions, aiming to advance automatic speech recognition (ASR) research focused on African accents in both clinical and general domains. This dataset contains 67,577 speech-transcript pairs, covering 120 distinct African native accents from 13 countries, with a total duration of 200.7 hours. During the dataset creation process, the research team collected text data from multiple African news and medical-related websites via web crawling and template-based techniques, to improve the dataset's diversity and relevance. AfriSpeech-200 has a wide range of application scenarios, and is designed to address the underrepresentation of African accents in ASR systems, especially in critical fields such as medical record processing, so as to enhance the efficiency and quality of medical services in Africa.