基于高斯混合模型的语音性别识别  

Speech gender recognition based on Gauss mixture model

作　　者：张超琼[1,2] 苗夺谦[1,2] 岳晓冬[1,2] 

机构地区：[1]同济大学计算机科学与技术系,上海201804 [2]同济大学嵌入式系统与服务计算教育部重点实验室,上海201804

出　　处：《计算机应用》2008年第S2期360-362,365,共4页journal of Computer Applications

基　　金：国家自然科学基金资助项目(6047501960775036);2006年博士学科点专项科研基金资助项目(20060247039)

摘　　要：利用高斯混合模型(GMM)方法进行语音的性别识别。首先概述了特征提取、识别方法及性别识别的过程;然后通过减少提取特征的语音帧数和降低高斯混合模型的混合阶数来提高性别识别速度;最后,将由M el频率倒谱参数(MFCC)特征和基音频率特征两种方法得到的测试样本后验概率结合,提出新的计算测试样本后验概率的方法。实验表明依据此后验概率能有效提高识别的正确率。In this paper,the method of Gauss Mixture Model(GMM) was used to do the gender identification.First,the method of feature extraction,gender identification and the recognition process are introduced.Second,the number of the frame in the feature extraction is reduced and the mixture number of the GMM was modified to improve the rate of the recognition.At last,the posterior probability of the feature Mel-Frequency Cepstral Coefficients(MFCC) was combined with the pitch result to form a new method for computing the posterior probability of the test example.The better correct rate of the recognition was improved by the experiment.

关 键 词：基音频率 高斯混合模型 性别识别 Mel频率倒谱参数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]