采用熵相关性优化分离性的SVM说话人识别

SVM speaker identification by adopting entropy correlation to optimize separation

针对目前说话人识别模型精度不高,应用性不强的缺点,提出一种采用熵相关性优化原始特征参数的方法,并综合特征熵相关性和原始特征特性值两方面因素改进了说话人识别的分离性测度。以说话人聚类类间差异最大化为目标,建立围绕基于特征分类相关性的参数自适应重构策略及分离性测度计算方法的说话人识别模型。仿真实验结果表明,该模型结构稳定,使说话人识别的精度及效率达到较好的平衡,具有较强的应用性能。

According to the accuracy is not high, and the application is not strong for speaker identification, a method by adopting entropy correlation to optimize original feature parameters is proposed, and the separation measure by combining feature entropy correlation and original feature value two factors is improved. The speaker identification model around the adaptive reconstruction strategy and the separation measure calculation based on feature classification correlation is proposed aimed to maximize differences between clas- ses speaker clustering. The simulation experiment shows that the proposed model has strong application performance with stable structure, and the accuracy and the speed achieve better balance for speaker identification.