Diffusion Deepfake Speech Dataset
收藏资源简介:
Diffusion Deepfake Speech Dataset是由布尔诺理工大学信息科学与技术学院创建的一个用于研究合成语音和深度伪造检测的数据集。该数据集包含183,400条由扩散模型生成的深度伪造语音记录,总时长约为336小时。数据集的创建使用了LJSpeech数据集,并通过扩散合成器和非扩散合成器生成。数据集的创建旨在评估扩散模型生成的语音对现有深度伪造检测系统的影响,并比较其与传统非扩散方法的语音质量。该数据集主要应用于合成语音和深度伪造检测的研究领域,旨在解决现有检测系统对新型合成语音的识别问题。
Diffusion Deepfake Speech Dataset is a dataset developed by the Faculty of Information Technology, Brno University of Technology for research on synthetic speech and deepfake detection. It contains 183,400 deepfake speech utterances generated by diffusion models, with a total duration of approximately 336 hours. The dataset was constructed using the LJSpeech dataset, and generated via both diffusion-based and non-diffusion-based speech synthesizers. It was created to evaluate the impact of diffusion model-generated speech on existing deepfake detection systems, and to compare the speech quality between diffusion-based and traditional non-diffusion-based synthesis methods. This dataset is primarily applied in the research fields of synthetic speech and deepfake detection, aiming to solve the problem where existing detection systems have difficulty identifying novel synthetic speech.




