Abstract: In this paper, the generalized networks based on ELM are proposed as recognition models.According to the experimental analysis, in the single hidden layer feed-forward network based on ELM, the multi-point crossover and multi-mutation genetic algorithm is used to optimize the parameters of the model.In the generalized single hidden layer feed-forward network based on kernel function ELM, grid search is used to find the optimal parameter combination of the model. On the basic of recognition results of the three emotions called happiness, anger and which are drawn from the TYUT and EMO-DB emotional speech databases. The proposed speech emotion recognition models based on ELM are superior to SVM on both generalization performance and speed.