论文部分内容阅读
针对支持向量机(SVM)输入参数不能充分利用高斯混合模型(GMM)均值、方差、权重所携带的说话人信息,而导致与文本无关话者确认系统性能下降的问题,本文结合GMM的均值、方差、权重,提出一种新的、基于自适应后GMM的,SVM模型输入特征提取方法。在NIST 06语音数据库上的实验表明,本方法将等误识率(EER)从高斯混合模型-通用背景模型(GMMUBM)系统的8.49%,下降到基于离散余弦变换(DCT)变换GMM-SVM系统的4.16%,以及基于主元成分分析(PCA)GMMSVM系统的3.3%.
Because the input parameters of SVM can not make full use of the speaker information carried by Gaussian mixture model (GMM) mean, variance and weight, it leads to the problem that the text independent of the speaker confirms the performance degradation of the system. In this paper, Variance and weight, a new feature extraction method based on adaptive GMM and SVM model input feature is proposed. Experiments on the NIST 06 voice database show that the proposed method reduces the equal error rate (EER) from 8.49% of the GMMUBM system to the GMM-SVM based on the discrete cosine transform (DCT) Of 4.16%, and 3.3% based on PCA GMMSVM system.