在噪声环境下的分级说话人辨识  

Hierarchical speaker identification under noisy environments

作　　者：邱政权[1] 尹俊勋[1] 杨俊[1] 

机构地区：[1]华南理工大学电子与信息学院,广州510640

出　　处：《控制与决策》2007年第5期581-584,共4页Control and Decision

基　　金：国家自然科学基金项目(60275005);广东省自然科学基金重点项目(04105938)

摘　　要：将小波变换与维纳滤波结合起来对语音进行去噪.为了提高系统的鲁棒性和辨识率,在采用分级说话人辨识的基础上,将基音周期的高斯概率密度对GMM分类器的似然度进行加权,形成新的似然度进行说话人辨识.实验结果显示,所提出系统的鲁棒性和辨识率都有所提高.Wavelet transform and Wiener filtering are combined to denoise speech. In order to improve the robustness and identification rate of the system, hierarchical speaker identification is proposed. Then the likeliness of GMM classifier is weighted by using the Gauss probability density of the pitch and novel likeliness is proposed for speaker identification. The experiment result shows that the robustness and the identification rate of the system proposed are both improved.

关 键 词：小波变换 维纳滤波 分级说话人辨识 基音周期 高斯概率密度 

分 类 号：TN912.34[电子电信—通信与信息系统]