To research the audio classification of “memory of the world heritage”Dongba classic books, the way of speech emotion feature extraction is adopted to identify the Dongba audio categories and realize the emotional state recognition of Dongba audio.And to improve the performance of human computer interaction, the way of speech emotion feature extraction based on Mel frequency cepstrum coefficient(MFCC) is adopted.The first order difference and the second order difference are introduced to describe the dynamic characteristics of speech features. The speech signal characteristic parameters based on MFCC and the shorttime energy are finally formed to extract the characteristics of speech emotion feature. The experiment research shows that the characteristics of the speech signal can be more clearly differentiate emotional information contained in the speech, and lay the foundation for recognizing speech emotion and the Dongba audio classification.