ADEPT

ADEPT数据集由Papercup Technologies Ltd.和爱丁堡大学创建，旨在评估语音合成中的韵律转移。该数据集包含552条自然语音样本，涵盖了情感、人际态度等全球韵律变化和话题强调、命题态度等局部韵律变化。数据集的创建过程涉及确定韵律类别、设计句子并录制，确保听众能准确区分不同韵律。ADEPT数据集主要应用于评估和改进文本到语音系统的韵律转移技术，解决韵律表达的自然性和多样性问题。The ADEPT dataset, developed by Papercup Technologies Ltd. and the University of Edinburgh, is constructed to evaluate prosody transfer in speech synthesis. It comprises 552 natural speech samples, covering global prosodic variations such as emotion and interpersonal attitude, as well as local prosodic variations including topic emphasis and propositional attitude. The development process of the dataset involved defining prosodic categories, designing target sentences and performing recordings, with measures taken to guarantee that listeners could accurately distinguish between different prosodic patterns. The ADEPT dataset is primarily utilized to assess and improve prosody transfer technologies in text-to-speech (TTS) systems, tackling the challenges related to the naturalness and diversity of prosodic expression.