大规模转换语音数据库
收藏资源简介:
本研究构建了一个名为‘大规模转换语音数据库’的数据集,由16种表现良好的语音转换方法生成,旨在推动源说话人验证任务的发展。数据集包含约327,600条转换语音样本,这些样本通过随机选择三个源语音样本对同一目标语音进行语音转换生成,模拟了三种不同的攻击者。创建过程中,采用了多种语音转换技术,确保数据集的多样性和实用性。该数据集主要应用于源说话人验证领域,旨在解决语音转换技术对自动说话人验证系统安全性的威胁问题,通过识别转换语音中的源说话人信息,提高系统的抗欺骗能力。
This study constructed a dataset named "Large-Scale Transformed Speech Database", which was generated by 16 well-performing speech conversion methods, aiming to advance the development of source speaker verification tasks. The dataset contains approximately 327,600 transformed speech samples, which are generated by randomly selecting three source speech samples to conduct speech conversion on the same target speech, simulating three distinct types of attackers. Multiple speech conversion technologies were adopted during the dataset creation process to ensure the diversity and practicality of the dataset. This dataset is primarily applied in the field of source speaker verification, aiming to address the security threats posed by speech conversion technologies to automatic speaker verification systems, and enhance the anti-deception capability of such systems by identifying the source speaker information within the transformed speech.




