Abstract: For Support Vector Machines (SVM) parameter optimization problem, we propose an improved bacterial foraging algorithm (Im-BFOA) and increase its learning ability in the practical speech emotion recognition. Firstly, we introduce simulated annealing (SA),Gaussian mutation and chaotic disturbance operator into BFOA to balance the efficiency of search and the diversity of population. Secondly, use Im-BFOA to optimize SVM parameters and propose a Im-BFOA-SVM method; Thirdly, based on prosodic features, quality features and chaotic features of speech, build a 144-dimension emotional feature vector and use FDR to dimension reduction to 5 dimensions; Finally, test the algorithm performance on the practical speech emotion database and compare the proposed algorithm Particle Swarm Optimization(PSO) algorithm to optimize the parameters of SVM (PSO-SVM method) with basic SVM methods and Back-Propagation (BP) neural network method. Experimental results show that average recognition rate of the Im-BFOA-SVM method reached 78.1%, respectively, higher than PSO-SVM method, SVM methods and BP neural network method of 3.7, 5.4 and 9.8%, indicating that Im-BFOA is a kind of effective SVM parameter selection method which can significantly improve practical speech emotion recognition rate.