摘要
提出了一种基于压缩感知(CS)的说话人识别算法以及在ARM系统中的实现,首先,介绍压缩感知理论框架,提出说话人识别可以与压缩感知理论相结合的依据;其次,提出基于压缩感知的说话人识别算法的基本方法,即建立说话人语音特征数据库和基追踪匹配得到最大均值系数,其中,语音特征向量由GMM均值超向量核算法得到,大量实验数据表明,该方法一定程度上提高了识别率,并且在说话人集合较大的情况下识别效果较好。
To improve the rates of speaker recognition,a method based on the compressed sensing（ CS） is proposed.First,the frame of compressed sensing theory is introduced to analyzing the premise of combining the compressed sensing theory with the speaker recognition. Then the major algorithm of speaker recognition based on compressed sensing is advanced,that is the establishment of speakers＇ characteristic database and matrix trace to obtain the maximum average coefficients matching. Oceans of experimental data indicate that this method has strong recognition ability and the performance is good when the collection of speakers is huge.
出处
《电子器件》
CAS
北大核心
2014年第6期1151-1154,共4页
Chinese Journal of Electron Devices
关键词
压缩感知
说话人识别
基追踪
高斯混合模型
compressed sensing
speaker recognition
matrix trace
Gaussian mixture model
