SceneFake
收藏资源简介:
SceneFake数据集是由中国科学院自动化研究所开发,专注于场景伪造音频检测。该数据集包含84,475条音频,通过使用语音增强技术仅改变真实语音的声学场景来生成伪造音频。数据集分为训练、开发、可见测试和不可见测试集,用于评估模型在检测未知场景伪造音频方面的性能。SceneFake数据集旨在解决音频伪造检测中的一个重要问题,即通过改变声学场景来伪造音频,这对于社会安全具有重大威胁。
The SceneFake dataset, developed by the Institute of Automation, Chinese Academy of Sciences, focuses on scene-forged audio detection. It contains 84,475 audio samples, where forged audios are generated by only modifying the acoustic scenes of genuine speech using speech enhancement techniques. The dataset is split into training, development, seen test, and unseen test sets, which are used to evaluate model performance in detecting unknown scene-forged audios. The SceneFake dataset aims to address a critical challenge in audio forgery detection: audio forgery achieved by altering acoustic scenes, which poses significant threats to social security.

- 1SceneFake: An Initial Dataset and Benchmarks for Scene Fake Audio Detection中国科学院自动化研究所 · 2024年



