论文部分内容阅读
应用正交实验设计方案对语音识别中特征参数的选择进行了计划、分析与实验设计,目的是在大量的特征参数中选择出具有互补作用的特征参数。该方法有4个特点:1)实验方案的构造方法简单,而且得到的实验方案具有各特征参数搭配均衡的特点;2)实验结果分析方法计算简便,计算结果的物理含意明显;3)只需很少的比较实验就可以找到识别性能较好的特征参数组合;4)已有的实验结果对后续实验方案的设计有很好的指导作用。实验结果表明正交实验设计用于特征参数选择是有效的。
Orthogonal experimental design is used to plan, analyze and design the selection of the characteristic parameters in speech recognition. The purpose is to select the characteristic parameters which have the complementary functions among a large number of characteristic parameters. The method has four characteristics: 1) the experimental scheme is simple, and the experimental scheme has the characteristics of balanced with each characteristic parameter; 2) the experimental results analysis method is simple, the physical meaning of the calculation results is obvious; 3) Very few comparative experiments can be found to identify better performance characteristics of the combination of parameters; 4) The existing experimental results on the design of the follow-up experimental program has a good guide. Experimental results show that orthogonal experimental design is effective for the selection of characteristic parameters.