Abstract: Speaker feature parameter extraction directly affects the establishment of recognition model. Parameter extraction methods of MFCC and LPC, respectively regard local low-frequency information and global AR signal as the main feature. In this paper, a method of speaker spectrum feature extraction based on duffing stochastic resonance is proposed. Simulation results show that this method, which provides a more elaborate recognition model for speaker recognition, can identify tiny differences of spectrum among speakers and extract basic characteristics of spectrum effectively.

Key words: duffing stochastic resonance, Speaker Recognition（SR）, feature parameter extraction

摘要： 说话人特征参数的提取直接影响识别模型的建立，MFCC与LPC参数提取方法，分别以局域低频信息和全局AR信号为主要特征。提出一种基于duffing随机共振的说话人频谱特征提取方法。仿真结果表明，该方法能识别说话人之间频谱的微小差别，有效地提取说话人频谱的基本特征，从而为说话人识别模型提供更为精细的识别模型。

关键词: duffing随机共振, 说话人识别（SR）, 特征提取