RFP数据集
收藏资源简介:
RFP数据集是由卡迪夫大学计算机科学与信息学院创建的一个包含真实、伪造和部分伪造音频的数据集,旨在帮助开发和评估伪造音频检测模型。该数据集包含五种不同的音频类型:部分伪造(PF)、带噪声的音频、语音转换(VC)、文本到语音(TTS)和真实音频。数据集通过结合多种音频源和技术生成,包括使用YouTube-8m等开放源项目获取真实音频,以及采用最新的TTS和VC技术生成伪造音频。此外,数据集还包括了127,862条来自354名不同年龄和地区的说话者的音频,用于评估检测模型的性能。RFP数据集的应用领域包括伪造音频检测、重放攻击检测和自动说话人验证系统等。
The RFP Dataset was developed by the School of Computer Science and Informatics at Cardiff University. It is a curated collection of genuine, forged, and partially forged audio samples, designed to support the development and benchmarking of forged audio detection models. The dataset comprises five distinct audio types: Partially Forged (PF), Noisy Audio, Voice Conversion (VC), Text-to-Speech (TTS), and Genuine Audio. It is constructed by integrating diverse audio sources and generation technologies: genuine audio is sourced from open-source projects such as YouTube-8m, while forged audio samples are generated using cutting-edge TTS and VC techniques. Furthermore, the dataset includes 127,862 audio clips from 354 speakers across various age groups and geographic regions, which serves as evaluation data for assessing the performance of audio detection models. Application domains of the RFP Dataset cover forged audio detection, replay attack detection, automatic speaker verification systems, and other related fields.




