SRC4VC

SRC4VC是由东京大学等机构创建的智能手机录制语音转换基准数据集，包含100名日本演讲者录制的11小时语音数据。数据集通过众包方式收集，每个演讲者录制了52个语音样本，分为四个子集：朗读、表情、对话和歌唱。数据集旨在评估语音转换模型在实际降级语音输入下的鲁棒性，特别关注于解决训练和评估数据之间的录制质量不匹配问题。

SRC4VC is a smartphone-recorded voice conversion benchmark dataset developed by the University of Tokyo and other institutions. It contains 11 hours of speech data recorded by 100 Japanese speakers. The dataset was collected via crowdsourcing, with each speaker recording 52 speech samples divided into four subsets: read speech, emotional speech, conversational speech, and singing. This dataset aims to evaluate the robustness of voice conversion models under real-world degraded speech inputs, with particular focus on addressing the recording quality mismatch between training and evaluation data.