基于SVM的藏语说话人性别识别

Tibetan Speaker Gender Recognition Based on SVM

说话人性别识别是语音识别研究中的一个重要分支.通过说话人的语音识别作为说话人性别识别的预分类技术可以降低研究问题的复杂度,提高系统的准确率.文中首先从建立的藏语语音性别库入手,提取语音的特征参数MFCC,进而利用SVM进行训练和识别.实验结果表明:用于说话人识别的MFCC特征能有效地用于藏语说话人性别识别,且与SVM联合可以得到比较好的效果,SVM的藏语说话人性别识别准确率达到了80%以上.

The Research on speaker gender recognition is an important branch of speech recognition,thus we can recognize speaker＇s gender by speaker＇s voice.Gender identification as a pre-classification of speaker recognition technology can lower the complexity of the problem and improve the system＇s accuracy.In this paper,firstly,we establish Tibetan speech gender library,extracting speech feature parameters MFCC,then we use SVM for training and recognition.The results show that：MFCC for speaker recognition features can be effectively used in Tibetan speaker gender identification,and with the combine of SVM can get better results,Tibetan speaker gender recognition accuracy rate achieved more than 80% by using SVM.