We consider the task of recognizing and learning the environments for mobile robot using audio information. Environments are mainly characterized by different types of specific sounds. Using audio enables the system to capture a semantically richer environment, as compared to using visual information alone. The goal of this paper is to investigate suitable features and the design feasibility of an acoustic environment recognition system. We performed statistical analysis of promising frequency- and time- domain based audio features. We show that even from unstructured environmental sounds, we can predict with fairly accurate results the type of environment that the robot is positioned.