Life Sounds Extraction and Classification in Noisy Environment

This paper deals with the sound event detection in a noisy environment and presents a first classification approach. Detection is the first step of our sound analysis system and is necessary to extract the significant sounds before ini-tiating the classification step. We present three original event detection algorithms. Among these algorithms, one is based on the wavelet and gives the best performances. We evaluate and compare their performance in a noisy en-vironment with the state of the art algorithms in the field. Then, we present a statistical study to obtain the acous-tical parameters necessary for the training and, the sound classification results. The detection algorithms and sound classification are applied to medical telemonitoring. We re-place video camera by microphones surveying life sounds in order to preserve patient's privacy.