Sounds wild

日期:2019-03-07 12:14:11 作者:荀崦媸 阅读:

By Matt Walker HOW do you count birds and animals in the wild when you can’t see them? A researcher in Philadelphia has the answer: he says a computer can locate hard-to-find creatures from their calls, even when they are hidden in dark rainforests or oceans. On an expedition to study birds in South America, John Spiesberger of the University of Pennsylvania found he could hear many more animals than he could see. The trouble was, he had no way of knowing exactly where their calls were coming from. So he decided to find a way to locate animals from their calls alone. At his home, Spiesberger set out an array of microphones pointing towards a patch of forest behind his house. He began trying to pinpoint birds there, using a signal-processing technique called cross-correlation which dates from Second World War radar systems. Any sound from a single source, such as a bird call, will reach two microphones at slightly different times (see Diagram). This time difference alone is not enough to show where the bird is, as the sound could have come from a whole range of places in that time interval. It turns out that you need five microphones to pinpoint the sound source in three dimensions—but there’s a snag. Cross-correlation is upset by wind, and also by variations in the local speed of sound through the air. And echoes from trees and rocks can produce false signals that scupper the computer’s ability to identify the correct signal and its location. But now, Spiesberger believes he has cracked the problem. He has harnessed another technique, called autocorrelation, which detects weak signals against strong background noise. This works by making weak signals constructively interfere, boosting them above background noise. Spiesberger found he could combine the two techniques: the cross-correlation function produces a range of locations for his backyard birds, while the autocorrelation function spots the correct location by searching for a telltale signature which indicates that the sound travelled straight to the microphones. Spiesberger has turned the combined algorithm into a program that runs on a laptop computer, and has successfully used it to pinpoint two red-winged blackbirds to within a metre. “I finally found a way to reliably locate calling animals,” he will tell a meeting of the Acoustical Society of America in Ohio next month. He expects his program will make it much easier to study populations of creatures in the wild. “It’s so hard to do this stuff in the dark, under water or in forests,