本征音自适应是一种快速自适应算法 ,它根据对说话人矢量全空间的本征分析指导参数更新。该文提出一种基于子空间分析的本征音自适应算法 ,并且不同于一般本征音自适应采用最大似然准则的做法 ,该算法用最大后验准则以更好地估计参数。实验证明 ,在仅有 1句自适应数据的情况下它即能取得 6 .4 5 %的相对误识率下降 ,自适应速度远快于传统的最大后验方法 ,也不存在最大似然线性回归方法在极少数据量情况下反而造成系统识别性能下降的现象。结果表明该方法并不明显依赖相关子空间的划分数量 ,是一种稳健的自适应方法 Eigenfrequency adaptive is a fast adaptive algorithm that guides parameter updates based on an eigenanalysis of the speaker’s full space. In this paper, an eigen-sound adaptive algorithm based on subspace analysis is proposed, which is different from that of the general eigenvoice adaptively adopting the maximum likelihood criterion. The algorithm uses the maximum a posteriori criterion to better estimate the parameters. Experiments show that with only one adaptive data, it can achieve a relative decline of 6.55%, adaptive speed much faster than the traditional maximum a posteriori method, there is no maximum likelihood linearity Regression method in the case of very little data instead of causing system identification performance degradation. The results show that this method does not depend significantly on the number of divisions in the relevant subspace and is a robust adaptive method