In order to improve the recognition rate of speech emotion recognition system

this paper studies a preferred method of supporting vector machine kernel function parameters. Firstly

according to the principle of support vector machine

the factors influencing the kernel parameters are given.Secondly

according to these factors

the kernel parameters of the support vector machine are optimized according to the Fisher criterion and the maximum entropy principle. Finally

five kinds of emotion recognition tests based on the emotional speech database are selected with the preferred parameters. The test results show that the fusion of the Fisher criterion and the maximum entropy method can effectively improve the accuracy of speech emotion recognition.