基于HMM的性别识别

Gender Identification Using HMM

进行男女生识别的方法有很多种,如GMM,VQ等,该文提出了基于HMM进行说话人性别识别的方法,该方法通过计算语音信号的Mel频率倒谱系数(MFCC)并使用隐马尔可夫模型(HMM)进行性别识别。在实验室环境下,对50个不同说话人(其中男女说话人各占一半)的语音文件采用该方法与基于VQ的方法进行比较实验,从实验方法和实验结果方面得出结论:HMM的方法更加简单易行,识别率更高。对于实验的语音材料,采用HMM的方法识别率可以达到100%。

There are many methods for gender identification,including GMM and VQ.This paper present s a method based on HMM.It calculates the MFCC on the voices to train two gender models and use s these two models to identify whether one speaker is male or female.Authors do experiments with50different speakers(25males and25females)using VQ and HMM separately in lab,and find that HMM is better than VQ,using the speech files,it gets a correct identification rate of100%.