基于修正Mel子带系数的文本无关的说话人识别

摘要

Text-independent speaker recognition is an important branch research field of speaker recognition because of its ease-to-use and potential applications in the information technology. This paper focuses on the problems of the feature extraction. Through modifying the Mel frequency subband coefficients,it enhances the differences of the frequency between the people in the system, improves the ability of separating the classes in the feature space. And the parameters can emphasize people's individuality and increase the average accuracy of recognition.%与文本无关的说话人识别具有用户使用方便、可应用范围较宽等优点,是当前说话人识别技术的研究重点.对文本无关说话人识别系统中的特征参数提取进行了研究,通过对Mel子带系数进行修正,增强了说话人识别系统中说话人之间的频带差异,提高了特征空间中类别的可分性,得到了更能体现说话人个性特征的Mel子带系数,从而提高了说话人识别系统的平均正确识别率.