基于GMM-UBM的语言辨识算法研究  

A Study of a Language Identification Algorithm Based on the GMM-UBM Model

作　　者：陈业仙[1] 张歆奕[1] 毛杰[1] 

机构地区：[1]五邑大学信息工程学院,广东江门529020

出　　处：《五邑大学学报（自然科学版）》2010年第3期56-60,共5页Journal of Wuyi University(Natural Science Edition)

摘　　要：运用Matlab软件,以自已建立的语音数据库为基础,对与文本无关的基于GMM-UBM的语言辨识系统进行了测试,获得的平均识别率达74%,与传统GMM算法的测试对比,基于GMM-UBM的语言辨识算法能更好地改善语言辨识系统的性能.Language identification technology is a very important part of the speech recognition technology.In this paper,based on the practical application and a self-established voice database,a language identification system based on the GMM-UBM model and independent of the speaker is studied and compared with the traditional GMM methods.Experiment results show that this algorithm can effectively improve the performance of the language identification system and achieve an average recognition rate of 74%.

关 键 词：语言辨识 高斯混合-全局背景模型 期望最大化 贝叶斯自适应算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]