With the spread of social needs and development of techniques, social interaction is more and more frequent among people. To promote and assist human social interaction, it's important to understand the social context the user situates. The paper mainly studies the understanding of social contexts based on background sounds, the goal of which is to recognize the social context in which users reside through analyzing the differences of background sounds. It uses the Mel frequency cepstral coefficients to analyze sound features and classify the sounds based on an improved Dynamic Time Warping (DTW) algorithm. Experimental results show that the proposed algorithm is more effective than traditional methods.