人的声音虽然具有个性特征,但随着电子信息技术的迅猛发展,人的语音也可以被模仿。当模仿者模仿说话人声音相似度极高时,当前说话人身份确认系统就有可能会被非法欺骗。研究了一种双模态信息融合的说话人识别方法,能把声纹信息和人脸信息融合在一起,通过动态贝叶斯网络(DBN)方法来刻画两类特征之间的相关性,并根据DBN模型作出最后的说话人识别结果。实验表明,基于双模态信息融合的说话人识别方法比传统单独的声纹识别和人脸识别有较高的正确识别率。
A bimodal information fusion method for speaker recognition is researched.The voiceprint information and face information are fused together.The correlation between the two types of features is characterized by DBN.Then,according to the DBN model the final result of speaker recognition is made.Experimental results show that the method of speaker recognition based on dual-mode information fusion has higher recognition rate than the traditional single voiceprint recognition and face recognition.