Detecting short-duration events from continuous sensor signals is a significant challenge in the domain of wearable devices and health monitoring systems. Time-series segmentation refers to the challenge of subdividing a continuous stream of data into discrete windows, which can be individually processed using statistical classifiers or other algorithms. In this paper, we propose an algorithm for segmenting time-series signals and detecting short-duration data in the domain of lightweight embedded systems with real-time constraints. First, we demonstrate an approach for signal segmentation using a simple binary classifier. Next, we show how a novel two-stage classification algorithm can reduce computational overhead compared to a single-stage approach. Our proposed scheme is benchmarked using an audio-based nutrition-monitoring case study.