Fake Speech Wild (FSW)

Fake Speech Wild (FSW)数据集是一个包含来自四个不同媒体平台（哔哩哔哩、YouTube、抖音和喜马拉雅）的254小时真实和深度伪造音频的中文数据集，专注于社交媒体。该数据集由北京邮电大学的研究人员创建，旨在解决深度伪造音频在社交媒体平台上日益严重的问题。数据集包含146,097个音频样本，总时长为254.58小时。数据集的创建过程包括人工收集、专家验证和语音活动检测，以确保音频样本的真实性和有效性。该数据集可用于评估深度伪造音频检测技术的性能，并提高其鲁棒性。

Fake Speech Wild (FSW) dataset is a Chinese-language social media-focused dataset that contains 254.58 hours of real and deepfake audio sourced from four distinct media platforms: Bilibili, YouTube, Douyin, and Himalaya. It was developed by researchers from Beijing University of Posts and Telecommunications to address the growing problem of deepfake audio on social media platforms. The dataset consists of 146,097 audio samples with a total duration of 254.58 hours. The creation of this dataset involves manual collection, expert verification, and voice activity detection to ensure the authenticity and validity of the audio samples. This dataset can be utilized to evaluate the performance of deepfake audio detection technologies and enhance their robustness.