一种新型的与文本相关的说话人识别方法研究

摘要

在与文本相关的说话人识别研究中,既要包含说话人身份的识别,又要包含语音文本内容的识别.提出一种基于语音识别的与文本相关的说话人识别方法,从而建立说话人的声纹模型和语音文本模型,与传统的仅建立一种模型的方法相比,该方法能更精确地描述说话人身份信息和语音的文本信息,较好地解决了短时语音样本识别效果不佳的问题.测试实验表明,和传统与文本相关的说话人识别方法(如基于动态时间规整、高斯混合-通用背景模型)相比,由本方法建立的系统虚警概率降低了8.9%,识别性能得到了提高.%In the study of text-related speaker recognition,it is to include the identity recognition as well as the speech text recognition.This paper proposes a new kind of text-related speaker recognition method based on the speech recognition.The model built by this method can describe both the identity information and the speech text information more accurately.Besides,it can also solve the problem that the short-term speech samples have poor recognition effect.The experiments show that compared with the traditional text-related speaker recognition system such as dynamic time warping(DTW) and Gaussian mixture model-universal background model(GMM-UBM),the false alarm probability of the system established by the present method is reduced by 8.9% and the recognition performance is improved.