论文部分内容阅读
支持向量机作为说话人建模方法用于与文本无关的话者确认研究时,如何提取适合SVM训练和测试的特征参数直接影响话者确认系统的性能和效率.根据高斯混合模型(GMM)聚类能力强的特点,提出一种基于自适应GMM聚类的说话人特征参数提取方法,通过自适应的GMM聚类将大样本、混叠严重的M FCC特征参数聚为小样本的、代表说话人个性特征的特征参数,并用于与文本无关的SVM话者确认.在N IST0′4 1side-1side数据库上的实验表明了该方法的有效性.
When SVM is used as a speaker modeling method to confirm textual research, how to extract feature parameters suitable for SVM training and test directly affects the performance and efficiency of speaker verification system.According to Gaussian mixture model (GMM) clustering This paper proposes a method for extracting speaker feature parameters based on adaptive GMM clustering. By using adaptive GMM clustering, large samples and severe aliasing M FCC feature parameters are gathered into small samples, representing the speaker Personality characteristics of the feature parameters and used for text-independent SVM speaker confirmation in the NST0’4 1side-1side database experiments show the effectiveness of the method.