Abstract: The pilots have various speaking styles in flight, such as emotional dialogue, speaking fast and slowly, speaking loudly and quietly, etc. Moreover, the physical and mental pressure of the pilots in flight can cause dialogue speech aberrance. If the speech was processed directly by the traditional speaker recognition system or a speech recognition system without any processing function, the speech performance will be poor. Therefore, the recognition of the pilots' speaking styles, a kind of paralinguistic information, was investigated to assist the subsequent speech recognition system and speaker recognition system. In the study, 6 925 samples were collected in the experiment database, 384-dimension acoustic features were extracted, and compared the classification ability of SVMs with different Kernel functions. The experiment results indicated that the SVM with Gauss radial basis Kernel function shows the best performance and its accuracy can reach 91.62%.