论文部分内容阅读
提出一种结合统计模型与区分性模型优点的说话人确认方法:基于GMM多维概率输出的SVM话者模型的说话人确认.以目标说话人的GMM模型对一条语音的不同特征分量的概率输出作为特征参数,建立目标说话人的SVM模型.在NIST’05 8conv4w-lconv4w数据库上的实验表明该方法的有效性.
A speaker verification method combining the advantages of statistical model and discriminative model is proposed: speaker verification of SVM speaker model based on GMM multi-dimensional probability output, probability output of different characteristic components of one speech as GMM model of target speaker Characteristic parameters, and establish the SVM model of the target speaker.Experiments on NIST’05 8conv4w-lconv4w database show the effectiveness of this method.