meituan/Audio-Turing-Test-Corpus

这是一个包含500个文本转录样本的多维中文语音转录数据集，用于评估机器生成语音的逼真度和自然度。数据集覆盖了中英混合、副语言特征和情感、特殊字符和数字、多音字以及古典中文诗歌/散文等五个语言和风格维度，并包括用于注意力检查的104个“陷阱”转录样本。

This is a multidimensional Chinese speech transcription dataset containing 500 textual transcripts, designed to evaluate the realism and naturalness of machine-generated speech. The dataset spans five linguistic and stylistic dimensions, including Chinese-English code-switching, paralinguistic features and emotions, special characters and numerals, polyphonic characters, and classical Chinese poetry/prose, and includes 104 trap transcripts for attentiveness checks during human evaluation.