首页> 外文会议>International Conference on Artificial Intelligence >SNR-multiconditon Approaches of Robust Speaker Model Compensation Based on PLDA in Practical Environment
【24h】

SNR-multiconditon Approaches of Robust Speaker Model Compensation Based on PLDA in Practical Environment

机译:基于PLDA在实践环境中基于PLDA的SNR-MulticonditOn方法

获取原文

摘要

In this paper, we focus on the research of robust speaker verification system in adverse noise conditions. To improve the performance of conventional single PLDA model in the practical environment, two SNR-multicondition strategies are adopted to solve the SNR mismatch problem. The first approach uses different PLDA models to match the relative SNR conditions of training and testing speech utterances. The other uses one PLDA model trained by speech utterances with different SNR and a multicondition score normalization method to adjust the score distribution. The performance of the two systems are demonstrated on a 120-speakers test set for each gender. According to the result of the experiment, the proposed approaches are proof to reduce the EER to 7.39%, 7.17% for male and 10.22%, 10.06% for female respectively in the random SNR noise condition, which means the SNR-multicondition systems are more robust in the real noisy environment.
机译:在本文中,我们专注于强大的扬声器验证系统在不良噪声条件下的研究。为了提高传统单层PLDA模型的实际环境的性能,采用了两个SNR多种情况策略来解决SNR不匹配问题。第一种方法使用不同的PLDA模型来匹配培训和测试语音话语的相对SNR条件。另一个使用具有不同SNR的语音话语训练的一个PLDA模型和多功能分数标准化方法来调整分数分布。两个系统的性能在每个性别的120扬声器测试装置上进行了演示。根据实验结果,拟议的方法证明,在随机SNR噪声条件下分别将eer和10.22%,10.22%,10.22%,10.06%的10.22%,即60.06%,这意味着SNR多种情况系统在真正的嘈杂环境中强大。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱:kefu@zhangqiaokeyan.com

京公网安备:11010802029741号 ICP备案号:京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有
  • 客服微信

  • 服务号