Automatically separating anthropogenic from natural sounds in parks.

Abstract

The National Park Service needs to determine the extent of both man-made and natural sounds in national parks. To do this, they need to separate anthropogenic noises from natural ones. To be accurate, this requires hundreds of hours of sound data spread out over times of day, days of the week, and seasons, and must be spatially sufficient. With these much data to analyze, manual listening for audible tones and noises can be an impossible task. However, with a program that uses an algorithm to search for anthropogenic sounds, the task becomes much easier. Our hypothesis is that all anthropogenic noises except for jet aircraft will include tones below 1000 Hz. Our main purpose is to write software that flags anthropogenic sounds below 1000 Hz. Because anthropogenic and natural sounds are uncorrelated, we hope to flag virtually all anthropogenic sounds at the cost of falsely flagging some natural sounds. Furthermore, we can very accurately estimate the ALEQ for a given length of time based on a much smaller sample and then subtract that from the total to find the ALEQ for the anthropogenic sound. This paper discusses the development and testing of this process.