基于高斯混合模型的说话人识别系统

针对概率得分均值法出现的单个帧概率打分容易畸低的情况以及投票法因归一化而损失掉正面影响帧的打分，提出了一种引入可信度的均值方法，实验证明：该方法兼顾二者的优势的同时，在一定程度上消除了各自产生的不利影响，提高了说话人识别的精度。

Speaker recognition system based on Gaussian mixture model

The paper proposes a probability average algorithm into which the creditable degree is introduced, aimed at preventing abnormal lowness from high occurrence in probability scoring of some frame in the way of average probability scoring and protecting some useful scores against loss in the algorithm of voting. It is proved that the algorithm not only combines the advantages of the two, but also enables the respective negative effects to be eliminated to a certain extent. Algorithm improves the precision of speaker recognition.