基于自适应RBFN阵列的说话人识别

提出一种可用于说话人识别的自适应RBFN阵列。RBF网设计的核心在于确定网络中心的数目及位置，该自适应算法有效地融合了IOC与ROLS算法的优点，不仅能动态调节RBF网的隐节点数，还能使网络的数据中心自适应变化，很好地优化了网络的结构。用与文本无关的闭集说话人识别系统对该算法进行了验证，实验结果表明，该方法与传统的RBF算法相比，自适应RBF网具有较好的鲁棒性以及精简的网络结构等优点。

Speaker Identification Using Adaptive RBF Networks Array

An efficient adaptive RBF networks array for speaker recognition is proposed in this paper.The key point in design of RBF networks is to specify the number and locations of the centers.The adaptive method combines the advantages of IOC and ROLS algorithms.Not only the number but also the positions of data centers are adapted in learning progress, which optimizes the structure of nets very well. Moreover,the experiments on a closed set, text-independent speaker recognition system show that, better robustness and simpler networks can be achieved through improved adaptive RBF networks in comparison with classical RBFN.