MLAAD
收藏资源简介:
MLAAD数据集是由弗劳恩霍夫AISEC创建的多语言音频反欺骗数据集,包含23种语言的76,000条合成语音数据。该数据集使用54种不同的TTS模型生成,总计163.9小时的语音数据,旨在解决现有反欺骗数据库主要集中于英语和中文,限制了其全球有效性的问题。MLAAD数据集通过提供多语言的合成语音,帮助训练和评估深度伪造检测模型,以提高其在实际应用中的性能。此外,该数据集的应用领域包括提高语音生物识别系统的安全性,以及对抗音频欺骗和深度伪造。
The MLAAD dataset is a multilingual audio anti-spoofing dataset developed by Fraunhofer AISEC. It contains 76,000 synthetic speech utterances across 23 languages, generated using 54 distinct TTS models with a total duration of 163.9 hours. This dataset was designed to address the limitation that existing anti-spoofing databases primarily focus on English and Chinese, which restricts their global applicability. The MLAAD dataset facilitates the training and evaluation of deepfake detection models by providing multilingual synthetic speech, thereby enhancing their performance in real-world applications. Furthermore, its application scenarios include improving the security of speech biometric systems and combating audio spoofing and deepfakes.




