mueller91/MLAAD

MLAAD（多语言音频反欺骗数据集）是一个用于训练、测试和评估音频deepfake检测的数据集。它基于M-AILABS数据集构建，包含了多种语言的伪造音频文件，每个语言目录下有不同模型生成的伪造音频，以及包含详细信息的元数据文件。该数据集旨在作为现有反欺骗模型的新的域外测试数据，或作为额外的训练资源。使用时建议结合M-AILABS中的真实音频文件，以获得平衡的数据集。

MLAAD: The Multi-Language Audio Anti-Spoofing Dataset is a dataset for training, testing, and evaluating audio deepfake detection. It is built upon the M-AILABS dataset and includes fake audio files in multiple languages, with each language directory containing audio files generated by different models, along with metadata files containing detailed information. The dataset is intended to be used as new out-of-domain test data for existing anti-spoofing models or as an additional training resource. It is recommended to complement the fake audios in MLAAD with the corresponding authentic ones from M-AILABS to obtain a balanced dataset.