一种GMM-SVM混合说话人辨认模型  

Speaker Identification Model Based on GMM-SVM

作　　者：冷自强[1] 王金明[2] 林大会[3] 

机构地区：[1]解放军理工大学通信工程学院研究生1队,江苏南京210007 [2]解放军理工大学通信工程学院电子信息工程系 [3]解放军理工大学通信工程学院研究生3队

出　　处：《军事通信技术》2009年第1期86-89,共4页Journal of Military Communications Technology

摘　　要：文中提出了一种GMM和SVM混合说话人识别模型,在特征参数域和概率得分域对两种模型进行了融合。混合模型结合了GMM和SVM各自的优势,使SVM的概率输出兼顾各说话人模型内部和模型之间的信息,并有效解决了SVM训练算法复杂,难以处理大量样本的问题。采用TIMIT数据库进行了说话人辨认实验,结果证明本文提出的GMM-SVM模型比传统的GMM模型和SVM模型具有更好的辨识性能。A hybrid speaker recognition model based on GMM and SVM was presented.GMM and SVM were mixed in both feature parameter and likelihood score domain.The new model combined the advantages of GMM and SVM,making the SVM output probabilities contain both the information inside and between the speaker models. The problem that SVM training algorithm is too complex to deal with large number of training data was resolved.The GMM-SVM model was tested on the TIMIT database and showed better performance than GMM and SVM...

关 键 词：说话人辨认 支持向量机 高斯混合模型 

分 类 号：TN912.34[电子电信—通信与信息系统]