机构地区: 北京大学深圳研究生院
出 处: 《华中科技大学学报(自然科学版)》 2013年第S1期184-187,共4页
摘 要: 提出了一种基于线性逻辑回归的方法,对利用基频和MFCC特征获得的分数进行融合来进行说话人的性别识别,其中包括了基于基频特征的单高斯模型和基于MFCC特征的混合高斯模型.采用语音库包括男性语音文件150个,女性语音文件190个.实验结果中识别率可高达97.65%,比传统单用基频或是MFCC特征的识别率都要高,具有更好的判别性能. A method based on linear logistic regression was proposed in this paper,which fused the scores for pitch and MFCC feature to discriminate the speaker′s gender.The voicebox for this paper included 150speech signals of male and 190speech signals of female.Experimental results show that the recognition rate can be as high as 97.65%,higher than that based on the traditional single fundamental frequency or MFCC feature,with better distinguish performance.