Parameters selection of support vector machine isa very important problem, which has great influence on itsperformance. In order to improve the learning andgeneralization ability of support vector machine, in thispaper, proposed a new algorithm -parallel artificial fishswarm algorithm to optimize kernel parameter and penaltyfactor of support vector machine, improved the loop body ofartificial fish swarm algorithm to avoid the missing of theoptimum solution, and proved its validity by testing withsome test functions; used the optimal parameters in a nonspecificpersons, isolated words, and medium-vocabularyspeech recognition system. The experimental results showthat the rates of speech recognition based on support vectormachine using the new algorithm are better than those ofusing the traditional artificial fish swarm algorithm indifferent signal to noise ratios and different words.Especially, the support vector machine model based on thenew algorithm can still maintain better recognition rates inlower signal to noise ratios. So the new algorithm is aneffective support vector machine parameter optimizationmethod, which makes the support vector machine not onlyhave good generalization ability, but have better robustness.
展开▼