用于说话人识别的基于可变因子整合的高斯混合模型  

Gaussian Mixture Model Based on Variable Factor-Integration for Speaker Recognition

作　　者：李杰[1] 刘贺平[1] 

机构地区：[1]北京科技大学信息工程学院,北京100083

出　　处：《模式识别与人工智能》2012年第6期937-942,共6页Pattern Recognition and Artificial Intelligence

基　　金：国家自然科学基金项目(No.10671011);北京市教委重点学科共建项目(No.XK100080537)资助

摘　　要：针对传统高斯混合模型在噪声环境下识别率明显下降的问题,在借鉴随机概率分布模型间的α因子融合机制基础上,提出基于可变因子α整合的高斯混合模型.该模型通过引入可变因子使得混合模型中不同成分所占的比重又得到一次调整.实验结果表明,通过对该模型参数进行重估计,在TIMIT/NTIMIT两种不同语料库和不同样本集的情况下识别率较传统高斯模型均有提高.尤其在噪声环境和α因子取最优值时,识别率可提高8%,在NIST评测数据集上与GMM-UBM系统对比,识别率也有提高.To solve the problem that the recognition rate of traditional Gaussian mixture model decreases significantly in noisy conditions, a Gaussian mixture model based on a variable factor-integration is presented by adopting the a-integration mechanism of multiple stochastic models in the form of probability distributions. Through introducing the variable factor, the proportion of different compositions in the mixture model is adjusted again. By re-estimating the proposed model parameters, the experimental results show the performance of the proposed model is better than that of the traditional Gaussian mixture model on databases TIMIT/NTIMIT and different speaker numbers. Especially in noisy conditions with the optimal value of a, the recognition rate is increased by 8%. On NIST evaluation database the experimental results show that the recognition rate is increased as well compared with GMM-UBM system.

关 键 词：可变因子 高斯混合模型 说话人识别 

分 类 号：TN912.34[电子电信—通信与信息系统]