Different inertia weight PSO algorithm optimizing SVM kernel parameters applied in a speech recognition system

Kernel parameters selection of support vector machine is a very important problem, which has great influence on the performance of support vector machine. In order to improve the learning and generalization ability of support vector machine and enhance speech recognition system accuracy, a method of searching for the Gaussian kernel support vector machine optimal parameters(C, ¿ ) based on particle swarm optimization is proposed in this paper. The inertia weight w, a key parameter of particle swarm optimization, is adopted three adjusting methods: w is constant 1, w changes according to linear decreasing and linear differential decreasing. This paper constructed a speech recognition system based on support vector machine using the optimized parameters. The speech data is isolated, non-specific and middle vocabulary words. The speech features are improved Mel-frequency cepstral coefficients. Experiments show that this method of particle swarm optimizing support vector machine kernel parameters is very efficient, in different SNRs and different words, all of the three w adjusting methods have higher speech recognition correct rates than default parameters of the support vector machine open source software, and linear decreasing inertia weight strategy has best recognition rates in most cases, which has practicability to some extent.