AS-70

该数据集名为AS-70，是首个公开可用的普通话口吃语音数据集，也是同类数据集中最大的一个，包含了对话和语音命令朗读的语音数据，且配有逐字的手动转录。除此之外，该数据集还包含了600条独特的语音命令，对五种具体类型的口吃事件进行了标注，是同类数据集中唯一一个非西方语言的开放数据集。在规模上，总共有来自70次录音会话的48.8小时语音数据。该数据集的任务涵盖了自动语音识别（ASR）和口吃事件检测（SED）。

This dataset, named AS-70, is the first publicly available Mandarin stuttering speech dataset and the largest of its kind. It contains speech data from both conversational speech and voice command recitations, paired with word-level manual transcriptions. Additionally, it includes 600 unique voice commands annotated with five specific types of stuttering events. AS-70 is the only open dataset for non-Western languages in its category. In terms of scale, it encompasses 48.8 hours of speech data from 70 recording sessions. The downstream tasks supported by this dataset cover Automatic Speech Recognition (ASR) and Stuttering Event Detection (SED).