Abstract: By using the ergodicity of chaos movement to improve the quality of the initial individual and Gaussian mutation, a modified Shuffled Frog Leaping Algorithm is proposed to increase the capacity of global search. Shuffled Frog Leaping Algorithm Neural Network，composed of modified Shuffled Frog Leaping Algorithm and neural network, is used in speech emotion recognition. According to dimensional model emotion features were extracted and categorized into prosody features and voice quality features. HNR feature was studied corresponding to different emotion categories. Modified Shuffled Frog Leaping Algorithm was used to train the random initial data, optimize the connection weights and thresholds of the neural network and have a fast network convergence speed. In the recognition experiments BP neural network, RBF neural network and Modified SFLA neural network were compared under the same testing environment. Modified SFLA neural network reached the highest recognition rate, 9.2% better than BP neural network and 7.9% better than RBF neural network. The results show that Modified Shuffled Frog Leaping Algorithm neural network brings a promising improvement in the classification ability for speech emotion recognition.