摘要
实现了一个基于高斯混合模型(GMM)的说话人辨识系统。GMM是用多个高斯分布的概率密度函数的组合来描述特征矢量在概率空间的分布状况,不同的说话人对应了不同的GMM。模型的训练采取了极大似然估计(ML)的EM方法。并在不同的数据集上实验,得到了好的结果。
This paper realized a speaker recognition system based on Gaussian Mixture Models （GMM）. GMM describe the probabilistic distribution of specific speaker＇s feature vectors by using mixed Gaussian pdfs. A Maximum Likelihood training method named E - M method has been applied in this system. Well performance of these experiments has been got in varied corous.
出处
《微处理机》
2006年第4期76-77,79,共3页
Microprocessors
