CFAD
收藏资源简介:
CFAD数据集是由中国科学院自动化研究所模式识别国家重点实验室创建的,用于研究假音频检测。该数据集包含347,400条音频,涵盖了12种主流语音生成技术生成的假音频,以及真实音频。为了模拟真实环境,数据集中的音频添加了三种噪声数据集的噪声,并在五种不同的信噪比下进行了处理。此外,还考虑了六种音频编码器进行音频转码。CFAD数据集不仅适用于假音频检测,还可用于音频取证中的假语音算法识别。该数据集的发布旨在推动假音频检测领域的进步,特别是在未知类型和复杂条件下的检测方法的泛化能力。
The CFAD dataset was developed by the State Key Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences for fake audio detection research. This dataset consists of 347,400 audio clips, including both fake audios generated by 12 mainstream speech synthesis technologies and authentic human audios. To simulate real-world environments, audios in the dataset are corrupted with noises from three noise datasets and processed under five different signal-to-noise ratios (SNRs). Additionally, six audio encoders are adopted for audio transcoding. The CFAD dataset is not only suitable for fake audio detection, but also can be applied to the recognition of fake speech algorithms in audio forensics. The release of this dataset aims to advance the development of the fake audio detection field, particularly the generalization capability of detection methods against unknown attack types and complex scenarios.

- 1CFAD: A Chinese Dataset for Fake Audio Detection中国科学院自动化研究所模式识别国家重点实验室 · 2023年



