Prevention of impostors entering speaker recognition systems
Prevention of impostors entering speaker recognition systems
复制标题
DOI：
--
复制
发表时间：
2008
期刊：
影响因子：
--
通讯作者：
Zheng Fang
中科院分区：
文献类型：
--
作者：
Zheng Fang
说话人识别系统需要防止冒名顶替者进入系统。高保真录音和说话人模拟是冒名顶替者最常用的两种方法。本文的重点是防止录音的使用。音频读取设备可以被视为一个模数转换系统，它会影响声音特征。可以分析这些特征来防止入侵。该模型基于高斯混合模型-通用背景模型(GMM-UBM)说话人识别系统，使用基于从语音数据采集的静音的语音信道模型。如果训练数据和测试数据的信道特征不同，则输入可能不可信。测试表明，该模型是基于高斯混合模型-通用背景模型(GMM-UBM)的说话人识别系统通过基于静音的信道检测，该系统与静音的等误率降低了约40%。
Speaker recognition systems need to guard against impostors entering the system.High-fidelity recording and speaker simulations are the two most common methods used by impostors.This paper focuses on preventing the use of recordings.Audio reading equipment can be regarded as an analog-to-digital conversion system which will affect the vocal characteristics.These characteristics can be analyzed to prevent incursions.The model is based on the Gaussian mixture model-universal background model(GMM-UBM) speaker recognition system using a voice channel model based on a mute voice gathered from voice data.If the channel characteristics for the training data and the test data differ,the input may not be trustworthy.Tests show that,with the channel examination based on the mute voice,the equal error rate of this system with the mute voice was reduced by about 40%.