Source Tracing of Audio Deepfake Systems: MLAAD Source Tracing Protocol

This data was created as part of the work entitled "Source Tracing of Audio Deepfake Systems", published in Interspeech 2024. In doing our training, development, and testing, Pindrop used WAV files from the Multi-Language Audio Anti-Spoof Dataset (MLAAD) and the M-AILABS Speech Dataset. You can obtain copies of these same WAV files directly from the developers here:MLAAD (version 1): https://owncloud.fraunhofer.de/index.php/s/tL2Y1FKrWiX4ZtP#editorM-AILABS: https://www.caito.de/2019/01/03/the-m-ailabs-speech-dataset/本数据集的构建基于发表于Interspeech 2024的研究成果《音频深度伪造系统溯源》（Source Tracing of Audio Deepfake Systems）。 在开展训练、开发与测试工作时，Pindrop团队使用了多语言音频反欺骗数据集（Multi-Language Audio Anti-Spoof Dataset，MLAAD）以及M-AILABS语音数据集的WAV格式音频文件。上述两份数据集的WAV文件均可通过以下链接直接从其开发者处获取： MLAAD（版本1）：https://owncloud.fraunhofer.de/index.php/s/tL2Y1FKrWiX4ZtP#editor M-AILABS：https://www.caito.de/2019/01/03/the-m-ailabs-speech-dataset/