LLM-Dys
收藏arXiv2025-09-30 收录
下载链接:
https://github.com/Berkeley-Speech-Group/LLM-Dys
下载链接
链接失效反馈官方服务:
资源简介:
该数据集是一个综合性的合成数据集,包含了11种言语不流畅的类型,总计含有12,790小时的言语不流畅语音。该数据集利用了大型语言模型增强的模拟技术,既捕捉了词汇层面也捕捉了音素层面的不流畅现象。规模上,该数据集达到了12,790小时,其任务是进行语音不流畅检测。
This is a comprehensive synthetic dataset that covers 11 types of speech disfluencies, with a total of 12,790 hours of disfluent speech audio. It leverages large language model-enhanced simulation techniques to capture disfluencies at both lexical and phonemic levels, and is designed for speech disfluency detection tasks.
提供机构:
Berkeley Speech Group



