24%。Animal's welfare can be feed back from their vocalism. Some experiments were conducted to get 35-day-old chicken's vocalization in different environments, including normal environment, light off environment and human disturbance environment. Based on Mel frequency cepstral coefficient (MFCC), a new algorithm for feature extraction——subband Mel spectrum centroid (SMSC) was adopted. Support vector machine（SVM）classification model was used to evaluate the performance of test using different sound parameters. The results show the accuracy of recognition of SMSC was 24% more than MFCC parameter.