VoxSim
收藏资源简介:
VoxSim是由韩国科学技术院和三星研究共同创建的大规模感知语音相似度数据集,包含约70,000个相似度评分,来自超过1,000名说话者。数据集内容主要来源于VoxCeleb1,涵盖多种声学环境和多样的说话者特征。创建过程中,通过随机生成语音对并进行听觉测试来收集评分。VoxSim主要应用于自动语音评估领域,旨在解决合成语音的说话者相似度评估问题。
VoxSim is a large-scale perceptual speech similarity dataset jointly developed by the Korea Advanced Institute of Science and Technology (KAIST) and Samsung Research. It contains approximately 70,000 similarity ratings collected from over 1,000 speakers. The dataset is primarily derived from VoxCeleb1, covering diverse acoustic environments and a wide range of speaker characteristics. During its creation, similarity ratings were collected by randomly generating speech pairs and conducting auditory listening tests. VoxSim is mainly applied in the field of automatic speech assessment, aiming to address the issue of speaker similarity evaluation for synthesized speech.




