Computer Engineering and Applications ›› 2011, Vol. 47 ›› Issue (10): 180-182.

• 图形、图像、模式识别 • Previous Articles     Next Articles

Speaker recognition method using combined features extraction and two-stage decision model

LI Zhanming,LIN Juan,CHEN Ruozhu   

  1. College of Electrical and Information Engineering,Lanzhou University of Technology,Lanzhou 730050,China
  • Received:1900-01-01 Revised:1900-01-01 Online:2011-04-01 Published:2011-04-01


李战明,林 娟,陈若珠   

  1. 兰州理工大学 电气工程与信息工程学院,兰州 730050

Abstract: Aimed at the problems that personalized feature extraction and impostor enter in speaker recognition,this paper proposes a speaker recognition method using combined features extraction and two-stage decision model.In the feature extraction stage,it uses combination of feature extraction combined with MFCC(Mel Frequency Cepstrum Coefficient) cepstrum feature,Delta_Delta feature and pitch extracting based on average amplitude difference method;in the recognition stage,it uses the score after score normalization compared with the unified threshold,and eliminates some impostor enter then combines two-stage decision model.Experimental results demonstrate the method improves the recognition rate.

Key words: speaker recognition, combined features extraction, score normalization, two-stage decision model

摘要: 针对目前说话人识别中个性化的特征提取以及假冒说话人的问题,提出一种组合特征提取和二级判断模型相结合的说话人识别方法。在特征提取阶段,采用MFCC倒谱特征、Delta_ Delta特征与平均幅度差法提取的基音周期相结合进行组合特征提取;在识别阶段,采用得分规整后的得分值与一个统一的阈值比较,将一部分假冒说话人排除后,再结合二级判断模型进行识别。实验结果证明该方法有效提高了识别率。

关键词: 说话人识别, 组合特征提取, 得分规整, 二级判断模型