Energy-efficient Ambient Sound Sensing and Classification Using Smart Phones

The sensing and classification for indoor and outdoor environment based on ambient sound, which is practical for the research and application of mobile computing, have gradually attracted the attention of researchers. At present, GPS and Wi-Fi are often used for location, however, the former can only be used for outdoor and the availability of the latter cannot be guaranteed at all times and all places. Hence we propose to take advantage of microphone of smart phone to sense and classify indoor and outdoor environment and try to find the tradeoff between classification accuracy and energy consumption through numerous experiments. This paper focuses on carrying on large number of experiment through using Samsung Nexus and collecting multiple sound samples of indoor and outdoor environment, then analyze and evaluate current advanced feature extraction methods for ambient sound and diversified classifiers, and get their performance of accuracy and energy consumption at different environment and different time slots in a day. After extensive experiment, we find that MFCC plus MP feature extraction method is best among others, and Naïve Bayes classifier can produce best classification accuracy while it is the most energy-intensive classifier. We summarize the most suitable value of the parameters, including 15s sample length and 256 frame length. In order to save more energy, except for reducing the frequency of use of Naïve Bayes classifier, we also can accelerate the speed of file I/O.