首页> 外文会议>International Speech Communication Association >Feature-Based and Channel-Based Analyses of Intrinsic Variability in Speaker Verification
【24h】

Feature-Based and Channel-Based Analyses of Intrinsic Variability in Speaker Verification

机译:基于特征和基于通道的扬声器验证的内部变异性分析

获取原文

摘要

We explore how intrinsic variations (those associated with thespeaker rather than the recording environment) affect text-independent speaker verification performance. In a previouspaper we introduced the SRI-FRTIV corpus and providedspeaker verification results using a Gaussian mixture model(GMM) system on telephone-channel speech. In this paper weexplore the use of other speaker verification systems on thetelephone channel data and compare against the GMMbaseline. We found the GMM system to be one of the morerobust across all conditions. Systems relying on recognitionhypotheses had a significant degradation in low vocal effortconditions. We also explore the use of the GMM system onseveral other channels. We found improved performance ontable-top microphones compared to the telephone channel infurtive conditions and gradual degradations as a function ofthe distance from the microphone to the speaker. Thereforedistant microphones further degrade the speaker verificationperformance due to intrinsic variability.
机译:我们探讨了内在的变化(与故障而不是录制环境相关联的那些)如何影响独立于文本的扬声器验证性能。以前的纸纸,我们在电话信道语音上使用高斯混合模型(GMM)系统推出了SRI-FRTIV语料库和提供的专业验证结果。在本文中,WEEXPLERE在Thetelephone通道数据上使用其他扬声器验证系统,并与GMMbaseline进行比较。我们发现GMM系统是跨所有条件的摩洛尔邦之一。依托识别内容的系统在低声努力下具有显着的降解。我们还探讨了GMM系统Onseveral其他频道的使用。我们发现,与电话频道诱导条件和逐渐降级的距离与麦克风到扬声器的距离的距离相比,我们发现了改进的性能。由于内在的可变性,因此,由于内在的变化,麦克风进一步降低了扬声器验证性能。

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