Audio was mainly used for speech and speaker recognition before. Sound event detection (SED) is another field of audio recognition which is the recognition of sounds other than speech and music. If we recognize environmental sounds coming from hazardous events then we can use this for surveillance for security. Audio surveillance can be integrated into video surveillance systems for public security in cities, for surveillance of elderly people living alone and road surveillance etc. In this paper we developed deep neural network (DNN) models to recognize scream and traffic accident (car crash). Our model tests show that the developed models can be used in real applications.