基于GMMs算法的说话人身份识别系统研究与实现  

Research and Implementation of Speaker Identification System Based on GMMs Algorithm

作　　者：赖万钦[1] 雷筱珍[2] LAI Wanqin;LEI Xiaozhen(College of Information and Intelligent Communications,Fujian Chuanzheng Communications College,Fuzhou,Fujian 350007,China;School of Innovation and Entrepreneurship,Fujian Chuanzheng Communications College,Fuzhou,Fujian 350007,China)

机构地区：[1]福建船政交通职业学院信息与智慧交通学院,福建福州350007 [2]福建船政交通职业学院创新创业学院,福建福州350007

出　　处：《闽江学院学报》2023年第2期31-40,共10页Journal of Minjiang University

基　　金：福建省中青年教师教育科研项目(科技类)(JAT201058)。

摘　　要：基于GMMs(Gaussian mixture models)的声纹识别算法,提出了针对声纹识别和静音检测的相应特征值提取方法。该方法通过采集、分析声音样本,提取其特征值,对特征值进行模拟训练并通过模型匹配来识别说话人身份。同时,提出了基于GMM模型聚类的VAD(voice activity detection)静音检测算法,该算法具备模型参数自适应功能,极大提高了系统对不同应用环境(嘈杂/安静)的适应能力。对比实验表明,所设计的说话人身份识别系统具有高效性、准确性。Based on the Gaussian mixture models(GMMs)algorithm,the corresponding eigenvalue extraction method for voice print recognition and silence detection is proposed.The method collects and analyzes sound samples to obtain eigenvalues,then simulate training on eigenvalues and identify the speaker′s identity through model matching.VAD silent detection algorithm based on GMM model clustering with model parameter adaptation function improves the adaptability of the system to different application environments(noisy/quiet).Comparative experiments show that speaker identification system so developed is efficient and accurate.

关 键 词：声纹识别 生物特征 GMMs算法 MFCC特征 静音检测算法 

分 类 号：R443.6[医药卫生—诊断学] TP212.9[医药卫生—临床医学]