Real-Time Seismic Event Detection Using Voice Activity Detection Techniques

Seismic event detection is a key element for volcano monitoring systems. Real-time event detection is required by early warning monitoring systems in order to minimize the possible impact of natural disasters of geophysical nature. In this paper, we propose to implement a real-time long period (LP) and a volcano-tectonic (VT) event detector based on voice activity detection algorithms. The main advantage of such detector is that it can also locate the endpoints of the seismic event. In order to determine the efficiency of the proposed detector, a database containing 436 seismic events (LP and VT) acquired from the Cotopaxi volcano in Ecuador was used for testing. Main performance parameters such as accuracy (A), precision, sensitivity or recall, specificity, and the balanced error rate (BER) were then calculated, finally the processing time required by the algorithm was also considered. The results obtained suggest comparable performance when compared to previously developed event detection algorithms for the same dataset but with much less computational complexity, achieving an A of 95.2% and a BER of 0.005. With further refinements the algorithm may provide a useful tool for real-time volcanic research.