balacoon/speech_gen_eval_testsets

该数据集包含用于评估语音生成系统性能的测试集。其中包含两个子集：vctk，是VCTK数据集的一个2k子集，具有标准的零样本语音生成评估功能；daps_celeb，是一个自定义数据集，结合了DAPS数据集的1.9k子集作为源音频和名人声音作为参考音频，模拟了将消费者麦克风（例如电话）录制的源语音转换成名人声音的使用场景。

This dataset comprises test sets designed to evaluate the performance of speech generation systems, including two subsets: vctk, a 2k subset of the VCTK dataset for standard zero-shot speech generation evaluation; and daps_celeb, a custom dataset that combines a 1.9k subset of the DAPS dataset as source audio with celebrity voices as reference audios, simulating a scenario where source speech recorded on consumer microphones (e.g., on a phone) is transformed into a celebrity voice using a random voice snippet from the internet.