基于贝叶斯网络的说话人识别研究  

Speaker recognition based on Bayesian network

作　　者：万洪杰[1] 杜利民[1] 邓浩江[1] 

机构地区：[1]中国科学院语音交互技术实验室

出　　处：《计算机应用》2005年第6期1334-1335,1338,共3页journal of Computer Applications

摘　　要：基于贝叶斯网络基础理论,并针对与文本无关说话人识别任务,构造了一种用于说话人识别的贝叶斯网络结构,给出了该网络模型的构造和参数估计方法,提出了进行说话人识别时进行推理和分类预测的算法。通过与传统的混合高斯模型(GMM)的实验类比,该贝叶斯网络模型能够在同样的训练和测试数据情况下识别率平均相对提高30%。Based on Bayesian network theory and according to the task of speaker recognition with text independent, a Bayesian network structure for speaker recognition was presented. Methods of constructing the network,parameter estimation and how to inference and classify in speaker recognition were also given in detail.Experimental results under the same train and test data show that this network structure can achieve thirty percent average relative improvement in performance than classical Gaussian mixture model.

关 键 词：说话人识别 文本无关 贝叶斯网络 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]