摘要
以高斯通用背景模型(Gaussian mixture model-universal background model,GMM-UBM)和i-vector模型为主的说话人识别算法在实际应用中取得了不错的成绩,但i-vector说话人识别模型中存在没有充分考虑通用背景(universal background,UB)数据与训练数据耦合性的问题导致模型性能不佳。提出了基于i-vector全局参数联合(global parameter joint of identify vector,GPJ-IV)的说话人识别方法。该方法利用背景说话人特征训练得到说话人通用背景模型(universal background model,UBM),构建基于全局联合差异空间和联合信道补偿的GPJ-IV模型。通过实验测试并与传统方法进行对比,实验结果显示,所提出的GPJ-IV模型相比i-vector模型,等错误率(equal error rate,EER)和最小检测代价函数(minimum detection cost function,MinDCF)性能分别提升了58.99%和15.9%。
In recent years,speaker recognition algorithms based on Gaussian mixture model-universal background Model(GMM-UBM)and i-vector Model(capacity is better than GMM-UBM)have developed and achieved good results in practical application.However,the coupling of universal background(UB)data with training data is not considered in speaker recognition of i-vector model.So,a speaker recognition algorithm based on Global parameter Joint of Identify Vector(GPJ-IV)is proposed.Firstly,the speaker universal background model(UBM)is obtained by using background speaker feature training.Secondly,this method uses background speaker feature training to obtain universal background model(UBM),and constructs GPJ-IV model based on global joint difference space and joint channel compensation.The speaker recognition test is carried out and compared with the traditional method.Experimental results show that the performance of the proposed GPJ-IV model is 58.99%and 15.9%higher than that of the I-vector model,respectively.
作者
杨明亮
龙华
邵玉斌
杜庆治
YANG Mingliang;LONG Hua;SHAO Yubin;DU Qingzhi(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,P.R.China)
出处
《重庆邮电大学学报（自然科学版）》
CSCD
北大核心
2021年第1期144-151,共8页
Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)
基金
国家地区自然科学基金(61761025)。
