基于TWSVMs的改进说话人识别系统

Improved Speaker Recognition System Based on Multi-class Twin Support Vector Machines

实现了一个基于双分界面的支持向量机的文本无关说话人识别系统,该系统在建立模型的过程中使用高斯混合模型进行特征提取,有效地减少了数据集的规模。与传统的支持向量机方法相比,该方法不仅达到了更高的识别率,对环境具有良好的鲁棒性,并且降低了算法的时间复杂度。由于该方法对大规模数据集的处理能力,使其比传统的方法更适合应用于实际。在相关实验中,也证实了该方法的有效性。

An improved speaker recognition system is implemented for text independent speaker recognition system with Twin Support Vector Machines （TWSVMs） and feature extraction based on Gaussian Mixture Models （GMMs）. The TWSVMS presents better performance than the standard SVMs, Because of the ability of processing large - scale dataset, the system is more suitable in the practice than traditional methods. The experimental results also prove the efficiency of the system.