在基于因子分析的说话人识别中,提出串行训练载荷矩阵的方法.在载荷矩阵训练中,采用串行的方式训练得到说话人因子矩阵、对角阵(残差矩阵)和信道空间矩阵.在说话人注册中,将以上3个载荷矩阵拼接,采用联合估计的方法得到每个说话人的因子.采用这种策略可有效解决因子分析中的饱和问题.在NISTSRE2006年核心测试数据库上等错误率能达到3.65%. In the speaker recognition based on factor analysis, a method of serial training load matrix is ​​proposed.In the training of load matrix, the speaker matrix, the diagonal matrix (residual matrix) and the channel space matrix are obtained by serial training. In the speaker registration, the above three load matrix splicing, the joint estimation of each speaker’s factor obtained by this strategy can effectively solve the problem of saturation in factor analysis in the NISTSRE2006 annual test data center The rate can reach 3.65%.