高斯序列核支持向量机用于说话人识别

Gaussian sequence kernel support vector machine for speaker recognition

说话人识别问题具有重要的理论价值和深远的实用意义,在研究支持向量机核方法理论的基础上,将其与传统高斯混合模型(GMM)相结合构建成基于高斯序列核的支持向量机(SVM)。SVM的灵活性和强大分类能力主要在于可以根据要处理的问题来相应的选取核函数。在识别的过程中引入特征空间归正技术NAP(Nuisance Attribute Projection)对同一说话人在不同信道和环境所带来的特征差异进行弥补。用美国国家标准与技术研究所(NIST)2004年评测数据集进行实验,结果表明该方法可以大幅度提高识别率。

Speaker recognition problems have important theoretical value and farreaching practical significance.On the basis of the support vector machine kernel methods,this paper combines it with traditional Gaussian Mixture Mode（lGMM） to build into a new support vector machine based on Gaussian sequence kernel.Much of the flexibility and classification power of SVM resides in the choice of kernel.And in the process of identifying,it introduces feature space norm technology performed by Nuisance Attribute Projection（NAP） to compensate the feature difference in different channels and environment from the same speaker.It is tested on the National Institute of Standards and Technology（NIST） 2004 evaluation database.Experiments results show that this method can greatly improve the recognition rate.