“Soft“ acoustic event detectors for limited resources platforms

Event detection algorithms cover enormous relevance when their execution environment has reduced capabilities as the CPU computing power, memory budget and overall energy capacity. It is the case for embedded wireless sensor nodes or motes that promise an actual and massive ubiquity on future real world applications. We have implemented a "soft" acoustic event detector of human activities, particularly speech signal. The algorithm has been adapted from the wireless multi-channel IEEE 802.11 QoS VoIP field. The events are detected on the base of the change of rate of the sum of the current input signal energy average plus the signal energy standard deviation values. The detector performance is presented in terms of false and positive detections, for three input signals with different SNR levels.