unfake/portufake

该数据集包含葡萄牙语音频深度伪造和真实说话者录音的频谱图，分别来源于Fake Voices Dataset和CETUC Corpus。数据集包含183,878个512px x 256px的彩色恒定Q变换（CQT）频谱图，分为“真实”或“伪造”两类标签。这些频谱图对应于91名男性和女性说话者的巴西葡萄牙语录音，以及从这些录音创建的深度伪造。录音部分来自CETUC Corpus，而深度伪造则取自Fake Voices Dataset。数据集旨在用于葡萄牙语音频深度伪造欺骗检测，特别是使用计算机视觉和图像分类方法。数据集结构包含一个.zip文件，其中有两个文件夹：“real”和“fake”，与“数据集描述”部分所述一致。

This dataset contains spectrograms of audio deepfakes and real speaker recordings in Portuguese, originating from Fake Voices Dataset and CETUC Corpus, respectively. The dataset contains 183,878 512px x 256px colored constant-Q transform (CQT) spectrograms categorized into two labels: real or fake. These correspond to Brazilian Portuguese recordings from 91 male and female speakers, and deepfakes created from those recordings. The recordings are part of CETUC Corpus, while the deepfakes were taken from Fake Voices Dataset. The dataset is intended to be used in audio deepfake spoofing in Portuguese language, using specially computer vision and image classification approaches. The dataset structure contains one .zip file, which has two folders in it: real and fake, according to what was described in the Dataset Description section.