In order to solve the problem of classification and recognition of vocal singing, this paper uses genetic algorithm (GA) to optimize the parameters of SVM, and then it uses the optimized SVM model for the classification of audio signals and recognition of common instruments, such as violin, cello, piano and flute. MATLAB simulation results show that the average recognition rate of GA-SVM model for piano, violin and flute is more than 92%, and the average recognition rate of GA-SVM model for cello is 52.1%. In addition, when utilizing GA-SVM model for the recognition of common instruments, the probability of misjudging cello as violin is 50.24%. Therefore, GA is adopted to improve SVM model. Compared with the SVM model, the average recognition rate of the improved SVM model increases significantly, and the probability of misjudgment decreases. This shows that the SVM model improved by GA has high accuracy for instrument audio signal recognition.